CN106161569A - The recommendation of Web content, buffer replacing method and equipment - Google Patents

Abstract

Embodiments of the present invention provide a network content recommendation and cache replacement method and device. The CDN sends the identifier of the cached content in the cached content queue to the recommender system, so that the recommender system will use the cached content sent by the CDN when recommending content to the user. The content generates recommendation results, so as to ensure that the recommendation results recommended by the recommendation system to the user are in the cache content queue of the CDN as much as possible, and do not need to be obtained from the original server where the content requested by the user is located, which reduces the waiting time of the user and can reduce CDN back-to-source bandwidth usage. The recommendation system will also generate the recommendation heat of all content based on the recommendation results, and send the recommendation heat of all content to the CDN, so that the CDN can take the recommendation heat into consideration when performing cache replacement, and keep the content with high recommendation heat in the CDN as much as possible. In the cache, users can get the content recommended by the recommendation system faster.

Description

Recommendation, cache replacement method and device for web content

technical field

The embodiment of the present invention relates to network content recommendation technology, and in particular to a method and device for network content recommendation and cache replacement.

Background technique

The emergence and popularization of the Internet has brought a large amount of information to users, which has met the needs of users for information in the information age. Sometimes you can't get the part of the information that is really useful to you, and the efficiency of using the information is reduced, which is the so-called information overload (information overload) problem. A very potential way to solve the problem of information overload is Recommendation Systems (referred to as RS). The recommendation system is used to recommend the information and products that the user is interested in to the user according to the user's information needs and interests.

Content distribution network (Content delivery network or Content distribution network, often abbreviated as CDN) is a distributed server system built on the Internet (Internet), which includes multiple data centers, and its purpose is to make network content faster, More stable transmission to users, reducing network delay and improving user experience. CDN nodes will be deployed in multiple locations and on multiple different networks, which can reduce the traffic required for transmission between users and network content, thereby reducing bandwidth costs, and CDN nodes will also dynamically transmit content to each other. And optimize the user's download behavior, thereby reducing the bandwidth cost required by the network content provider, improving the user's download speed, and improving the stability of the system.

In the existing network system, the recommendation system and the CDN system work independently, which may cause the content recommended by the recommendation system to not be in the CDN cache, so that when the user requests network content from the CDN edge server, the CDN needs to go back to the source to obtain the network content. That is, the CDN edge server needs to first obtain the network content from the server where the network content is located in the CDN cache, and then, the CDN edge server sends the network content to the user, and the CDN returns to the source to obtain the network content, which will cause the user to wait for a long time and occupy the CDN Back-to-source bandwidth, etc.

Contents of the invention

Embodiments of the present invention provide a network content recommendation and cache replacement method and device, which have solved the problem that the content recommended by the recommendation system to the user is not in the CDN cache, resulting in long waiting time for the user.

The first aspect of the present invention provides a method for recommending network content, including:

The recommendation system receives the identifier of the cached content in the cached content queue sent by the content distribution network CDN, obtains the information of the cached content according to the identifier of the cached content and the entire content library, and combines the identifier of the cached content with the cached content The information added to the cache content library;

When the recommendation system receives the recommendation request message sent by the client, the recommendation system uses the first recommendation algorithm to calculate the first recommendation result according to the pre-acquired user interest characteristics and the entire content library;

The recommendation system acquires a second recommendation result according to the cached content library;

The recommendation system fuses the first recommendation result and the second recommendation result according to a preset fusion algorithm to obtain a target recommendation result;

The recommendation system pushes the target recommendation result to the target user.

With reference to the first aspect, in a first possible implementation manner of the first aspect, the recommendation system obtains the second recommendation result according to the cached content library, including:

The recommendation system calculates the second recommendation result by using a second recommendation algorithm according to the user interest characteristics and the cached content library.

With reference to the first aspect, in a second possible implementation of the first aspect, the recommendation system acquires a second recommendation result according to the cached content library, including:

The recommendation system selects recommended content belonging to the cached content library from the first recommendation result, and takes the selected recommended content as the second recommendation result.

In combination with the first aspect, any one of the first to second possible implementations of the first aspect, in the third possible implementation of the first aspect, the recommendation system integrates The algorithm fuses the first recommendation result and the second recommendation result to obtain a target recommendation result, including:

The recommendation system determines common recommendation content in the first recommendation result and the second recommendation result;

The recommendation system deletes the common recommendation content from the first recommendation result to obtain a third recommendation result;

The recommendation system uniformly sorts the recommended content in the second recommendation result and the third recommendation result according to the score of the recommended content;

The recommendation system takes the sorted recommended content as the target recommendation result, or the recommendation system selects part of the recommended content from the sorted recommended content according to a preset algorithm as the target recommendation result.

In combination with the first aspect, any one of the first to second possible implementations of the first aspect, in the fourth possible implementation of the first aspect, the recommendation system integrates The algorithm fuses the first recommendation result and the second recommendation result to obtain a target recommendation result, including:

The recommendation system determines common recommendation content in the first recommendation result and the second recommendation result;

The recommendation system deletes the common recommendation content from the first recommendation result to obtain a third recommendation result;

The recommendation system selects a%*k recommended content from the third recommendation result, where k is the number of recommended content included in the target recommendation result, and a is greater than or equal to 0 and less than or equal to 100;

The recommendation system selects (1-a%)*k recommended content from the second recommendation result;

The recommendation system uniformly sorts the recommended content selected from the third recommendation result and the recommended content selected from the second recommendation result according to the scores of the recommended content, and uses the unified sorted recommended content as the Target recommendation results.

With reference to the fourth possible implementation of the first aspect, in a fifth possible implementation of the first aspect, the recommendation system selects a%*k recommended content from the third recommendation result, including:

The recommendation system sorts the recommended content in the third recommendation result according to the score of the recommended content, and selects the top a%*k recommended content from the sorted third recommendation result;

The recommendation system selects (1-a%)*k recommended content from the second recommendation result, including:

The recommendation system sorts the recommended content in the second recommendation result according to the score of the recommended content, and selects the top (1-a%)*k recommended content from the sorted second recommendation result .

With reference to any one of the third to fifth possible implementations of the first aspect, in a sixth possible implementation of the first aspect, the recommendation system deletes from the first recommendation results After the common recommended content, the method also includes:

The recommendation system increases the score of the common recommended content included in the second recommendation result.

In combination with the first aspect, any one of the first to second possible implementations of the first aspect, in the seventh possible implementation of the first aspect, the recommendation system integrates The algorithm fuses the first recommendation result and the second recommendation result to obtain a target recommendation result, including:

The recommendation system determines common recommendation content in the first recommendation result and the second recommendation result;

The recommendation system deletes the common recommendation content from the second recommendation result to obtain a fourth recommendation result;

The recommendation system uniformly sorts the recommended content in the first recommendation result and the fourth recommendation result according to the score of the recommended content;

The recommendation system takes the sorted recommended content as the target recommendation result, or the recommendation system selects part of the recommended content from the sorted recommended content according to a preset algorithm as the target recommendation result.

In combination with the first aspect, any one of the first to second possible implementations of the first aspect, in the eighth possible implementation of the first aspect, the recommendation system integrates The algorithm fuses the first recommendation result and the second recommendation result to obtain a target recommendation result, including:

The recommendation system determines common recommendation content in the first recommendation result and the second recommendation result;

The recommendation system deletes the common recommendation content from the second recommendation result to obtain a fourth recommendation result;

The recommendation system selects a%*k recommended content from the first recommendation result, where k is the number of recommended content included in the target recommendation result, and a is greater than or equal to 0 and less than or equal to 100;

The recommendation system selects (1-a%)*k recommended content from the fourth recommendation result;

The recommendation system uniformly sorts the recommended content selected from the first recommendation result and the recommended content selected from the fourth recommendation result according to the scores of the recommended content, and uses the unified sorted recommended content as the recommended content. The target recommendation results.

With reference to the eighth possible implementation of the first aspect, in the ninth possible implementation of the first aspect, the recommendation system selects a%*k recommended content from the first recommendation result, including:

The recommendation system sorts the recommended content in the first recommendation result according to the score of the recommended content, and selects the top a%*k recommended content from the sorted first recommendation result;

The recommendation system selects (1-a%)*k recommended content from the fourth recommendation result, including:

The recommendation system sorts the recommended content in the fourth recommendation result according to the score of the recommended content, and selects (1-a%)*k recommended content from the sorted fourth recommendation result.

With reference to any one of the seventh to ninth possible implementations of the first aspect, in a tenth possible implementation of the first aspect, the recommendation system deletes from the second recommendation results After the common recommended content, the method also includes:

The recommendation system increases the score of the common recommended content included in the first recommendation result.

In combination with the first aspect, any one of the first to second possible implementations of the first aspect, in the eleventh possible implementation of the first aspect, the method further includes:

The recommendation system generates a recommendation popularity library according to the recommendation situation of all the contents in the entire content library, and the recommendation popularity library includes the recommendation popularity of all the contents in the entire content library within a preset time;

The recommendation system sends all the content in the recommended popularity library to the CDN.

With reference to the eleventh possible implementation manner of the first aspect, in a twelfth possible implementation manner of the first aspect, the recommendation system performs a calculation of the first recommendation result and the second recommendation result according to a preset fusion algorithm. The two recommendation results are fused, and after the target recommendation result is obtained, the method further includes:

The recommendation system updates the recommendation popularity library according to the target recommendation result.

In combination with the first aspect and any one of the first to twelfth possible implementations of the first aspect, in the thirteenth possible implementation of the first aspect, the cached content sent by the CDN It is the content of the top P% of the cache content queue, or the incremental data of the content of the top P% of the cache content queue relative to the content sent last time, where P is greater than 0 and less than 100.

The second aspect of the present invention provides a cache replacement method for network content, including:

The content distribution network CDN edge server obtains the recommendation popularity and access popularity of the cached content in the cached content queue;

The CDN edge server performs cache replacement on the cache content queue according to the access popularity and recommendation popularity of the cache content in the cache content queue.

With reference to the second aspect, in a first possible implementation manner of the second aspect, the CDN edge server performs cache replacement on the cache content queue according to the access popularity and recommendation popularity of the cache content in the cache content queue, include:

If the size of the cached content in the cached content queue is greater than or equal to the first threshold, the CDN edge server determines to eliminate the cached content at the end of the cached content queue with less popular access;

The CDN edge server compares the recommended popularity of the cached content with the same access popularity at the tail of the cache content queue, and eliminates the cache content with the lower recommended popularity among the cache contents with the same access popularity, until the cache content queue If the size of the cached content in the queue is smaller than a second threshold, the elimination of the cached content queue is stopped, and the second threshold is smaller than or equal to the first threshold.

With reference to the second aspect, in a second possible implementation manner of the second aspect, the CDN edge server performs cache replacement on the cached content queue according to the access popularity and recommendation popularity of the cached content in the cached content queue, include:

If the size of the cached content in the cached content queue is greater than or equal to the first threshold, the CDN edge server determines to eliminate the cached content at the queue tail of the cached content queue with less popular access;

The CDN edge server calculates the comprehensive popularity of the cache content at the tail of the cache content queue according to the access popularity and recommendation popularity of the cache content at the tail of the cache content queue;

The CDN edge server eliminates cached content with low overall popularity in the tail of the cached content queue until the size of the cached content in the cached content queue is smaller than a second threshold, then stops eliminating the cached content queue , the second threshold is less than or equal to the first threshold.

With reference to the second aspect, in a third possible implementation manner of the second aspect, the CDN edge server obtains the recommendation popularity and access popularity of the cached content in the cached content queue, including:

The CDN edge server generates the access heat of the cached content in the cached content queue according to the historical access conditions of the cached content in the cached content queue;

The CDN edge server receives the recommendation heat of the cache content in the cache content queue sent by the recommendation system, and the recommendation heat of the cache content in the cache content queue is the recommendation system according to the cache content in the cache content queue generated by the recommendations.

In combination with the second aspect, any one of the first to third possible implementations of the second aspect, in a fourth possible implementation of the second aspect, the method further includes:

The CDN edge server acquires recommendation popularity and access popularity of candidate content in the candidate content queue;

The CDN edge server performs cache replacement on the candidate content queue according to the recommendation popularity and access popularity of the candidate content in the candidate content queue.

With reference to the fourth possible implementation manner of the second aspect, in a fifth possible implementation manner of the second aspect, the CDN edge server evaluates the The above candidate content queues are used for cache replacement, including:

If the size of the candidate content in the candidate content queue is greater than or equal to a third threshold, the CDN edge server determines to eliminate the candidate content at the tail of the candidate content queue that has less popular access;

The CDN edge server compares the recommendation popularity of the candidate content with the same access popularity in the candidate content queue, and eliminates the candidate content with the lower recommendation popularity among the candidate content with the same access popularity, until the candidate content in the candidate content queue If the size of the candidate content is smaller than a fourth threshold, the elimination of the candidate content queue is stopped, and the fourth threshold is smaller than or equal to the third threshold.

With reference to the fourth possible implementation manner of the second aspect, in a sixth possible implementation manner of the second aspect, the CDN edge server evaluates the The above candidate content queues are used for cache replacement, including:

If the size of the candidate content in the candidate content queue is greater than or equal to a third threshold, the CDN edge server determines to eliminate the candidate content at the tail of the candidate content queue that has less popular access;

The CDN edge server calculates the comprehensive popularity of the candidate content in the queue tail of the candidate content queue according to the access popularity and recommendation popularity of the candidate content in the queue tail of the candidate content queue;

The CDN edge server eliminates the candidate content with lower comprehensive popularity of the candidate content in the tail of the candidate content queue until the size of the candidate content in the candidate content queue is smaller than the fourth threshold, then stops processing the candidate content The queue is eliminated, and the fourth threshold is less than or equal to the third threshold.

With reference to the fourth possible implementation manner of the second aspect, in the seventh possible implementation manner of the second aspect, the CDN edge server obtains the recommendation popularity and access popularity of the candidate content in the candidate content queue, including:

The CDN edge server generates the access heat of the candidate content in the candidate content queue according to the historical access conditions of the candidate content in the candidate content queue;

The recommendation system receives the recommendation popularity of the candidate content sent by the recommendation system, and the recommendation popularity of the candidate content is generated by the recommendation system according to the recommendation situation of the candidate content.

A third aspect of the present invention provides a cache replacement method for network content, including:

The edge server of the CDN receives the content acquisition request sent by the client, and the content acquisition request includes identification information of the content to be accessed;

The CDN edge server determines whether the content to be accessed is in its cache content queue according to the identification information of the content to be accessed;

If the content to be accessed is in the cached content queue, the CDN edge server returns the content to be accessed to the client;

The CDN edge server updates the access popularity of the content to be accessed, and calculates the popularity information of the content to be accessed according to the access popularity and recommendation popularity of the content to be accessed, and updates the Cache content queue;

When it is necessary to perform cache replacement on the cached content queue, the CDN edge server eliminates less popular cached content in the cached content queue according to the popularity information of the cached content in the cached content queue.

With reference to the third aspect, in a first possible implementation manner of the third aspect, if the identification information of the content to be accessed is not in the cached content queue, the CDN edge server Determine whether the content to be accessed is in the candidate content queue of the CDN edge server;

If the content to be accessed is in the candidate content queue, the CDN edge server updates the access popularity of the content to be accessed, and determines the popularity of the content to be accessed according to the access popularity and recommendation popularity of the content to be accessed information;

The CDN edge server determines whether the popularity of the content to be accessed is greater than a preset popularity threshold according to the popularity information of the content to be accessed;

If the popularity of the content to be accessed is greater than the popularity threshold, the CDN edge server adds the content to be accessed to the cache content queue, and deletes the content to be accessed from the candidate content queue;

The CDN edge server returns to the client the IP address of the original server where the content server to be accessed is located.

With reference to the third aspect, in a second possible implementation manner of the third aspect, if the content to be accessed is not in the candidate content queue, the CDN edge server adds the content to be accessed to the candidate content in the queue;

The CDN edge server updates the access popularity of the content to be accessed, determines the popularity information of the content to be accessed according to the popularity of the content to be accessed and the recommendation popularity of the content to be accessed, and determines the popularity information of the content to be accessed according to the popularity of the content to be accessed information to update the candidate content queue;

The CDN edge server returns to the client the IP address of the original server where the content server to be accessed is located.

With reference to the third aspect, in a third possible implementation manner of the third aspect, if the popularity of the content to be accessed is less than or equal to the popularity threshold, the CDN edge server, according to the popularity information of the content to be accessed updating the candidate content queue;

The CDN edge server returns to the client the IP address of the original server where the content server to be accessed is located.

In combination with the third aspect and the first to third possible implementations of the third aspect, in a fourth possible implementation of the third aspect, the method further includes:

When it is necessary to cache and replace the candidate content queue, the CDN edge server eliminates less popular candidate content in the candidate content queue according to the popularity information of the candidate content in the candidate content queue.

A fourth aspect of the present invention provides a recommendation system, including:

The receiving module is configured to receive the identifier of the cached content in the cached content queue sent by the content distribution network CDN, obtain the information of the cached content according to the identifier of the cached content and the entire content library, and combine the identifier of the cached content with the Add the information about the cached content to the cached content repository;

A recommendation module, configured to use a first recommendation algorithm to calculate a first recommendation result according to the pre-acquired user interest characteristics and the entire content library when the recommendation system receives a recommendation request message sent by the client;

The recommendation module is further configured to obtain a second recommendation result according to the cached content library;

a fusion module, configured to fuse the first recommendation result and the second recommendation result according to a preset fusion algorithm to obtain a target recommendation result;

A sending module, configured to push the target recommendation result to the target user.

With reference to the fourth aspect, in a first possible implementation manner of the fourth aspect, the recommendation module acquires a second recommendation result according to the cached content library, specifically:

According to the user interest feature and the cached content library, the second recommendation result is calculated by using a second recommendation algorithm.

With reference to the fourth aspect, in a second possible implementation manner of the fourth aspect, the recommendation module acquires a second recommendation result according to the cached content library, specifically:

Select recommended content belonging to the cached content library from the first recommended result, and use the selected recommended content as the second recommended result.

In combination with the fourth aspect, any one of the first to second possible implementations of the fourth aspect, in the third possible implementation of the fourth aspect, the fusion module is specifically used for:

determining common recommended content in the first recommendation result and the second recommendation result;

deleting the common recommendation content from the first recommendation result to obtain a third recommendation result;

According to the score of the recommended content, the recommended content in the second recommended result and the third recommended result is uniformly sorted;

The sorted recommended content is used as the target recommendation result, or part of the recommended content is selected from the sorted recommended content according to a preset algorithm as the target recommendation result.

In combination with the fourth aspect, any one of the first to second possible implementations of the fourth aspect, in the fourth possible implementation of the fourth aspect, the fusion module is specifically used for:

determining common recommended content in the first recommendation result and the second recommendation result;

deleting the common recommendation content from the first recommendation result to obtain a third recommendation result;

Select a%*k recommended content from the third recommendation result, where k is the number of recommended content included in the target recommendation result, and a is greater than or equal to 0 and less than or equal to 100;

Select (1-a%)*k recommended content from the second recommended result;

According to the scores of the recommended content, the recommended content selected from the third recommended result and the recommended content selected from the second recommended result are uniformly sorted, and the uniformly sorted recommended content is used as the target recommended result.

With reference to the fourth possible implementation of the fourth aspect, in a fifth possible implementation of the fourth aspect, the fusion module selects a%*k recommended content from the third recommendation result, specifically :

sorting the recommended content in the third recommendation result according to the score of the recommended content, and selecting the top a%*k recommended content from the sorted third recommendation result;

The fusion module selects (1-a%)*k recommended content from the second recommendation result, specifically:

The recommended content in the second recommended result is sorted according to the score of the recommended content, and the top (1-a%)*k recommended content is selected from the sorted second recommended result.

With reference to any one of the third to fifth possible implementations of the fourth aspect, in a sixth possible implementation of the fourth aspect, the fusion module deletes from the first recommendation result Following the common recommended content, it is also used to:

The score of the common recommended content included in the second recommendation result is increased.

In combination with the fourth aspect, any one of the first to second possible implementations of the fourth aspect, in the seventh possible implementation of the fourth aspect, the fusion module is specifically used to:

determining common recommended content in the first recommendation result and the second recommendation result;

deleting the common recommendation content from the second recommendation result to obtain a fourth recommendation result;

According to the score of the recommended content, the recommended content in the first recommended result and the fourth recommended result is uniformly sorted;

The sorted recommended content is used as the target recommendation result, or part of the recommended content is selected from the sorted recommended content according to a preset algorithm as the target recommendation result.

In combination with the fourth aspect, any one of the first to second possible implementations of the fourth aspect, in the eighth possible implementation of the fourth aspect, the fusion module is specifically used for:

determining common recommended content in the first recommendation result and the second recommendation result;

deleting the common recommendation content from the second recommendation result to obtain a fourth recommendation result;

Select a%*k recommended content from the first recommendation result, where k is the number of recommended content included in the target recommendation result, and a is greater than or equal to 0 and less than or equal to 100;

Select (1-a%)*k recommended content from the fourth recommended result;

According to the score of the recommended content, the recommended content selected from the first recommended result and the recommended content selected from the fourth recommended result are uniformly sorted, and the unified sorted recommended content is used as the target recommended result. .

With reference to the eighth possible implementation of the fourth aspect, in the ninth possible implementation of the fourth aspect, the fusion module selects a%*k recommended content from the first recommendation result, specifically :

sorting the recommended content in the first recommendation result according to the score of the recommended content, and selecting the top a%*k recommended content from the sorted first recommendation result;

The fusion module selects (1-a%)*k recommended content from the fourth recommendation result, specifically:

The recommended content in the fourth recommended result is sorted according to the score of the recommended content, and (1-a%)*k recommended content is selected from the sorted fourth recommended result.

With reference to any one of the seventh to ninth possible implementations of the fourth aspect, in a tenth possible implementation of the fourth aspect, the fusion module deletes Following the common recommended content, it is also used to:

The score of the common recommended content included in the first recommendation result is increased.

In combination with the fourth aspect, any one of the first to second possible implementations of the fourth aspect, in the eleventh possible implementation of the fourth aspect, the recommendation system further includes:

A recommendation popularity generation module, configured to generate a recommendation popularity library according to the recommendation situation of all contents in the entire content library, the recommendation popularity library including the recommendation popularity of all the contents in the entire content library within a preset time;

The sending module is further configured to send all the content in the recommended popularity library to the CDN.

With reference to the eleventh possible implementation manner of the fourth aspect, in a twelfth possible implementation manner of the fourth aspect, the fusion module performs the calculation of the first recommendation result and the second recommendation result according to a preset fusion algorithm. The two recommendation results are fused, and after the target recommendation result is obtained, the recommendation heat generation module is also used for:

The recommendation popularity database is updated according to the target recommendation result.

In combination with the fourth aspect and any one of the first to twelfth possible implementations of the fourth aspect, in the thirteenth possible implementation of the fourth aspect, the cached content sent by the CDN It is the content of the top P% of the cache content queue, or the incremental data of the content of the top P% of the cache content queue relative to the content sent last time, where P is greater than 0 and less than 100.

A fifth aspect of the present invention provides a content distribution network CDN edge server, including:

An acquisition module, configured to acquire recommendation popularity and access popularity of cached content in the cached content queue;

A cache replacement module, configured to perform cache replacement on the cache content queue according to the access popularity and recommendation popularity of the cache content in the cache content queue.

With reference to the fifth aspect, in a first possible implementation manner of the fifth aspect, the cache replacement module is specifically configured to:

If the size of the cached content in the cached content queue is greater than or equal to the first threshold, it is determined to eliminate the cached content at the end of the cached content queue with less popular access;

Comparing the recommendation heat of the cache content with the same access heat at the queue tail of the cache content queue, eliminating the cache content with the lower recommendation heat among the cache content with the same access heat until the cache content in the cache content queue reaches If the size is smaller than a second threshold, stop eliminating the cached content queue, and the second threshold is smaller than or equal to the first threshold.

With reference to the fifth aspect, in a second possible implementation manner of the fifth aspect, the cache replacement module is specifically configured to:

If the size of the cached content in the cached content queue is greater than or equal to the first threshold, it is determined to eliminate the cached content at the queue tail of the cached content queue with less popular access;

According to the access heat and recommendation heat of the cache content at the queue tail of the cache content queue, calculate the comprehensive heat of the cache content in the queue tail of the cache content queue;

Eliminate the cache content with less comprehensive heat in the queue tail of the cache content queue until the size of the cache content in the cache content queue is less than a second threshold, then stop eliminating the cache content queue, the second The threshold is less than or equal to the first threshold.

With reference to the fifth aspect, in a third possible implementation manner of the fifth aspect, the obtaining module is specifically configured to:

generating the access heat of the cached content in the cached content queue according to the historical access conditions of the cached content in the cached content queue;

receiving the recommendation heat of the cache content in the cache content queue sent by the recommendation system, the recommendation heat of the cache content in the cache content queue is generated by the recommendation system according to the recommendation situation of the cache content in the cache content queue .

In combination with the fifth aspect and any one of the first to third possible implementations of the fifth aspect, in the fourth possible implementation of the fifth aspect, the acquisition module is further configured to:

Obtain the recommendation popularity and access popularity of the candidate content in the candidate content queue;

The cache replacement module is further configured to: perform cache replacement on the candidate content queue according to the recommendation popularity and access popularity of the candidate content in the candidate content queue.

With reference to the fourth possible implementation of the fifth aspect, in the fifth possible implementation of the fifth aspect, the cache replacement module is specifically configured to:

If the size of the candidate content in the candidate content queue is greater than or equal to a third threshold, it is determined to eliminate the candidate content at the tail of the candidate content queue that has less popular access;

Comparing the recommendation popularity of the candidate content with the same access popularity in the candidate content queue, eliminating the candidate content with the lower recommendation popularity among the candidate contents with the same access popularity, until the size of the candidate content in the candidate content queue is less than the fourth threshold, stop eliminating the candidate content queue, and the fourth threshold is less than or equal to the third threshold.

With reference to the fourth possible implementation of the fifth aspect, in a sixth possible implementation of the fifth aspect, the cache replacement module is specifically configured to:

If the size of the candidate content in the candidate content queue is greater than or equal to a third threshold, it is determined to eliminate the candidate content at the tail of the candidate content queue that has less popular access;

calculating the comprehensive popularity of the candidate content in the tail of the candidate content queue according to the access popularity and recommendation popularity of the candidate content in the queue tail of the candidate content queue;

Eliminate the candidate content with less comprehensive popularity of the candidate content in the queue tail of the candidate content queue, until the size of the candidate content in the candidate content queue is less than the fourth threshold, then stop eliminating the candidate content queue, so The fourth threshold is less than or equal to the third threshold.

With reference to the fourth possible implementation of the fifth aspect, in a seventh possible implementation of the fifth aspect, the acquiring module is specifically configured to:

generating the access popularity of the candidate content in the candidate content queue according to the historical access conditions of the candidate content in the candidate content queue;

The recommendation popularity of the candidate content sent by the recommendation system is received, and the recommendation popularity of the candidate content is generated by the recommendation system according to the recommendation situation of the candidate content.

A sixth aspect of the present invention provides an edge server of a content distribution network CDN, including:

A receiving module, configured to receive a content acquisition request sent by the client, where the content acquisition request includes identification information of the content to be accessed;

A processing module, configured to determine whether the content to be accessed is in its own cache content queue according to the identification information of the content to be accessed, and return to the client if the content to be accessed is in the cache content queue The content to be accessed;

An update module, configured to update the access popularity of the content to be accessed, and calculate the popularity information of the content to be accessed according to the access popularity and recommendation popularity of the content to be accessed, and update the Cache content queue;

The cache replacement module is configured to, when it is necessary to perform cache replacement on the cache content queue, eliminate less popular cache content in the cache content queue according to the popularity information of the cache content in the cache content queue.

With reference to the sixth aspect, in the first possible implementation manner of the sixth aspect, if the identification information of the content to be accessed is not in the cached content queue, the processing module is further configured to:

determining whether the content to be accessed is in the candidate content queue of the CDN edge server according to the identification information of the content to be accessed;

If the content to be accessed is in the candidate content queue, update the access popularity of the content to be accessed, and determine the popularity information of the content to be accessed according to the access popularity and recommendation popularity of the content to be accessed;

judging whether the popularity of the content to be accessed is greater than a preset popularity threshold according to the popularity information of the content to be accessed;

If the popularity of the content to be accessed is greater than the popularity threshold, adding the content to be accessed to the cache content queue, and deleting the content to be accessed from the candidate content queue;

Returning to the client the IP address of the original server where the content server to be accessed is located.

With reference to the sixth aspect, in a second possible implementation manner of the sixth aspect, if the content to be accessed is not in the candidate content queue, the processing module is further configured to:

adding the content to be accessed to the candidate content queue;

Update the access popularity of the content to be accessed, determine the popularity information of the content to be accessed according to the popularity of the content to be accessed and the recommendation popularity of the content to be accessed, and update the candidate according to the popularity information of the content to be accessed content queue;

Returning to the client the IP address of the original server where the content server to be accessed is located.

With reference to the sixth aspect, in a third possible implementation manner of the sixth aspect, if the popularity of the content to be accessed is less than or equal to the popularity threshold, the processing module is further configured to:

updating the candidate content queue according to the popularity information of the content to be accessed;

Returning to the client the IP address of the original server where the content server to be accessed is located.

With reference to the sixth aspect and the first to third possible implementations of the sixth aspect, in a fourth possible implementation of the sixth aspect, the cache replacement module is further configured to:

When it is necessary to cache and replace the candidate content queue, the less popular candidate content in the candidate content queue is eliminated according to the popularity information of the candidate content in the candidate content queue.

In the network content recommendation and cache replacement method and device provided by the embodiments of the present invention, the CDN sends the identifier of the cached content in the cached content queue to the recommender system, so that the recommender system will use the cached content sent by the CDN when recommending content to the user. Generate recommendation results to ensure that the recommendation results recommended by the recommendation system to users are in the cached content queue of the CDN as much as possible. When the user requests the recommended content included in the recommended results from the CDN, the CDN can obtain the user request from the cached content queue The content does not need to be obtained from the original server where the content requested by the user is located, which reduces the waiting time of the user and can reduce the bandwidth usage of the CDN back to the source. The recommendation system will also generate the recommendation heat of all content based on the recommendation results, and send the recommendation heat of all content to the CDN, so that the CDN can take the recommendation heat into consideration when performing cache replacement, and keep the content with high recommendation heat in the CDN as much as possible. In the cache, users can get the content recommended by the recommendation system faster.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description These are some embodiments of the present invention. For those skilled in the art, other drawings can also be obtained according to these drawings on the premise of not paying creative efforts.

Fig. 1 is a schematic diagram of an existing video recommendation process;

FIG. 2 is a flowchart of a method for recommending network content provided by Embodiment 1 of the present invention;

FIG. 3 is a schematic structural diagram of a CDN and a recommendation system provided by Embodiment 2 of the present invention;

4 is a flow chart of the CDN sending the identifier of cached content to the recommendation system provided by Embodiment 2 of the present invention;

FIG. 5 is a flow chart of sending the recommendation popularity of the entire content library to the CDN by the recommendation system provided by Embodiment 3 of the present invention;

FIG. 6 is a schematic diagram of a business process of a method for recommending network content provided in Embodiment 4 of the present invention;

FIG. 7 is a flowchart of a method for recommending network content provided in Embodiment 4 of the present invention;

FIG. 8 is a flowchart of a method for recommending network content provided in Embodiment 5 of the present invention;

9 is a schematic structural diagram of a cache content queue and a candidate content queue of a CDN edge server;

FIG. 10 is a flowchart of a cache replacement method for network content provided in Embodiment 6 of the present invention;

FIG. 11 is a flow chart of a cache replacement method for network content provided by Embodiment 7 of the present invention;

FIG. 12 is a flowchart of a cache replacement method for network content provided in Embodiment 8 of the present invention;

FIG. 13 is a flowchart of a cache replacement method for network content provided by Embodiment 9 of the present invention;

FIG. 14 is a schematic structural diagram of a recommendation system provided by Embodiment 10 of the present invention;

FIG. 15 is a schematic structural diagram of a CDN edge server provided by Embodiment 11 of the present invention;

FIG. 16 is a schematic structural diagram of a CDN edge server provided by Embodiment 13 of the present invention;

FIG. 17 is a schematic structural diagram of a CDN edge server provided by Embodiment 14 of the present invention;

FIG. 18 is a schematic structural diagram of a CDN edge server provided by Embodiment 15 of the present invention;

FIG. 19 is a schematic structural diagram of a recommendation system provided by Embodiment 16 of the present invention.

detailed description

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is a part of embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

At present, many network application systems include both CDN and recommendation system, such as Internet Protocol Television (IPTV for short), Video On Demand (VoD for short), video websites (such as Netflix, YouTube, Youku and LeTV, etc.), e-commerce (Electronic Commerce) websites (such as Taobao, Jingdong and Amazon, etc.), mobile application and game stores (such as Apple's App Store), etc.

Recommendation systems can be divided into three categories according to recommendation algorithms: content-based recommendation systems, collaboration-based recommendation systems, and hybrid recommendation systems. The content-based recommendation system obtains user interest feature vectors and content feature vectors, and recommends items for users by calculating the similarity between user interest feature vectors and content feature vectors. Collaboration-based recommendation systems do not rely on recommending users and content itself, but recommend items for target users based on items purchased by other users. A hybrid recommender system uses both content-based and collaboration-based recommendations. In general, the purpose of recommendation is to select the most useful items for users from the massive content library and present them to users, so as to prevent users from getting lost in a wide variety of choices, save users' time, and increase the exposure of target products . It should be noted that the content mentioned in the following embodiments is also referred to as network content, and the network content may specifically be videos, music, commodities, applications (APPs), games, and the like.

The CDN includes multiple network clusters. The network clusters included in the CDN can be divided into central clusters and edge clusters. The central cluster includes one or more CDN central servers, and the edge cluster includes one or more CDN edge servers. The central cluster generally caches all network content that needs to be distributed, while the edge cluster only caches a part of the network content that needs to be distributed. When a user accesses a certain content, the user sends an address resolution request to the Domain Name System (DNS) through the client, and the DNS will assign the user's access to the edge cluster closest to the user. The content accessed by the user, then the CDN edge server storing the content accessed by the user in the edge cluster will directly return the content requested by the user to the user. If the edge cluster accessed by the user does not store the content accessed by the user, the edge cluster will obtain the accessed content from the original server where the accessed content is located, and then return it to the user. In the latter case, for users, users often need to wait for a long time to get a response from the edge cluster, which affects user experience. Back-to-origin bandwidth leads to a waste of network resources. The back-to-origin bandwidth is the bandwidth between the CDN and the original server of the content.

Taking video recommendation as an example, FIG. 1 is a schematic diagram of an existing video recommendation process. As shown in FIG. 1 , the video recommendation process may include the following steps:

Step 101, the client sends a video recommendation request message to a web server (Web Server).

In video recommendation, the client can be a browser or a player. This process is generally triggered by the user. For example, to open the Youku player, click the movie option on the home page of the player, and the Youku player will automatically send the video to the web server. Send a video recommendation request message.

Step 102, the web server sends a query request message to the recommendation system.

The query request is used to obtain the query recommendation list. The recommendation system receives the message after the video recommendation request sent by the web server, obtains user interest characteristics, and obtains recommendation results from the video content library according to the preset recommendation algorithm and user interest characteristics. Among them, the user interest characteristics are mainly generated based on the following two types of information: one type of information is the information provided by the user when logging into the web server, including: the user's gender, age, region, education background, personal interests, etc., and the other type of information is the user's Historical access records. The recommendation system can be located in the same hardware device as the web server, or in a separate hardware device.

Step 103, the recommendation system returns a video recommendation list to the web server.

After obtaining the recommendation result, the recommendation system can sort the videos included in the recommendation result according to the video scores to obtain a video recommendation list, and return the video recommendation list to the web server.

Step 104, the web server returns a video recommendation list to the client.

Step 105, the client requests a certain video content from the CDN edge server.

This operation is generally triggered by the user clicking (or watching) a certain video in the recommended video list.

Step 106, the CDN edge server returns the requested video content to the client.

The video requested by the client may or may not be cached in the CDN edge server. When the video requested by the client is on the CDN edge server, the CDN edge server will directly send the video requested by the client to the client. , if the video requested by the client is not on the CDN edge server, the CDN edge server will request the video from the original server where the video is located, cache the video in the CDN edge server, and then the CDN edge server will return the video to the client .

From the above examples, we can see that when the content recommended by the recommendation system to the user is not in the CDN, the CDN needs to go back to the source to obtain the network content, which will cause the user to wait for a long time and occupy the CDN back-to-source bandwidth.

In order to solve the problems of the prior art, Embodiment 1 of the present invention provides a method for recommending network content. When the recommendation system recommends content to users, it will consider the content cached in the CDN, and ensure that the content recommended to users is stored in the CDN. FIG. 2 is a flowchart of a method for recommending network content provided in Embodiment 1 of the present invention. As shown in FIG. 2 , the method provided in this embodiment may include the following steps:

Step 201: The recommendation system receives the identifier of the cached content in the cached content queue sent by the CDN, obtains the information of the cached content according to the identifier of the cached content and the entire content library, and adds the identifier of the cached content and the information of the cached content to the cached content library.

The CDN sends the identifier of the cached content stored in the cached content queue to the recommendation system under the trigger of the trigger signal. The cached content’s identifier can be a uniform/universal resource locator (url for short, or url of the cached content) of the cached content. The hash value (HASH). The cache content queue of CDN is used to store the cache content cached in the CDN. When the CDN sends the identifier of the cache content to the recommendation system, the identifiers of all the cache content in the cache content queue can be sent to The recommendation system can also only send part of the cached content in the cache queue. The main reason is that the cached content in the CDN cached content queue is arranged in the order of popularity, and the cached content in the cached content queue is when the cached content queue is full. Will be eliminated from the cache content queue, if the recommender system sends the cached content that may be eliminated in the cached content queue to the recommender system, and the recommender system recommends the cached content to the user, when the user accesses the cached content, the cached The content may have been eliminated. At this time, the CDN still needs to obtain the cached content from the original server that stores the cached content.

In order to ensure that the cached content in the cached content queue recommended by the recommendation system is still in the cached content queue when the user accesses it, in this embodiment, the cached content sent by the CDN is the content of the top P% of the cached content queue, or the cached Incremental data of the content of the top P% of the content queue relative to the content sent last time, where P is greater than 0 and less than 100.

After the recommendation system receives the identifier of the cached content, it generates a cached content library according to the cached content ID and the entire content library. Metadata, content score and other information, taking video content as an example, the metadata of video content includes content attribute tags, and content attribute tags include title, author, actor, style, shooting time, etc. Taking product content as an example, product The attribute tags of content include commodity name, price, store name, color, style, etc. After the CDN only sends the identifier of the cached content to the recommendation system, the recommendation system finds the information of the cached content from the entire content library according to the identifier of the cached content, and adds the information of the cached content and the identifier of the cached content to the cached content library. The cached content library is A subset of the overall content library.

Step 202. When the recommendation system receives the recommendation request message sent by the client, the recommendation system uses the first recommendation algorithm to calculate the first recommendation result according to the pre-acquired user interest characteristics and the entire content library.

All recommended content in the first recommendation result belongs to the entire content library. During the recommendation process, the recommendation system will generate the characteristics of the entire content library according to the characteristics of the entire content library, and generate the first recommendation result according to the user interest characteristics and the characteristics of the entire content library.

Step 203, the recommendation system obtains the second recommendation result according to the cached content library.

In an implementation manner, the recommendation system uses a second recommendation algorithm to calculate and obtain a second recommendation result according to user interest characteristics and the cached content library. In another implementation manner, the recommendation system selects recommended content belonging to the cached content library from the first recommendation result, and takes the selected recommended content as the second recommendation result. In the former implementation manner, the recommendation system generates the characteristics of the cached content library according to the cached content library, and generates the second recommendation result according to the user interest characteristics and the characteristics of the cached content library. In the latter implementation, the recommendation system first determines which recommended content belongs to the cache content library from the first recommendation result, and then obtains the recommended content belonging to the cache content library from the first recommendation result, and uses the selected recommended content as The second recommended result.

In this embodiment and the following embodiments, the features of user interests, features of the entire content library, and features of the cached content library mentioned are feature vectors.

In this embodiment, both the first recommendation algorithm and the second recommendation algorithm may include one or more recommendation algorithms, and the recommendation algorithm may adopt any existing algorithm, for example, a content-based algorithm, a collaboration-based algorithm, a A hybrid algorithm of content and collaboration. And in this embodiment, the first recommendation algorithm and the second recommendation algorithm may be the same or different.

In this embodiment, both the first recommendation result and the second recommendation result include at least one recommendation content, and the first recommendation result and the second recommendation result may be represented in a list form. It should be noted that step 202 and step 203 are executed in no sequence, and may also be executed simultaneously.

Step 204, the recommendation system fuses the first recommendation result and the second recommendation result according to a preset fusion algorithm to obtain a target recommendation result.

Specifically, the first recommendation result and the second recommendation result can be fused in the following ways:

In the first way, first, the recommendation system determines the common recommendation content in the first recommendation result and the second recommendation result. Secondly, the recommendation system deletes the common recommendation content from the first recommendation result to obtain the third recommendation result; then, the recommendation system uniformly sorts the recommendation content in the second recommendation result and the third recommendation result according to the score of the recommendation content; Finally, the recommendation system takes the sorted recommendation content as the target recommendation result, or, the recommendation system selects part of the recommendation content from the sorted recommendation content according to the preset algorithm as the target recommendation result. For example, the recommendation system needs to return a total of K recommendations content, the recommendation system selects the top K content from the sorted recommended content as the target recommendation result.

Both the first recommendation result and the second recommendation result include multiple recommended contents, and the recommended contents included in the first recommendation result and the second recommendation result may be partly the same. The recommendation system compares the first recommendation result and the second recommendation result determine the common recommended content in both.

In the second way, first, the recommendation system determines the common recommendation content in the first recommendation result and the second recommendation result, deletes the common recommendation content from the first recommendation result, and obtains the third recommendation result. Secondly, the recommendation system selects a%*k recommended content from the third recommendation result, where K is the number of recommended content included in the target recommendation result, a is greater than or equal to 0 and less than or equal to 100, and selects from the second recommendation result (1-a%)*k recommended content. Finally, the recommendation system uniformly sorts the recommended content selected from the third recommendation result and the recommended content selected from the second recommendation result according to the scores of the recommended content, and takes the unified ranked recommended content as the target recommendation result.

The recommendation system selects a%*k recommended content from the third recommendation result. Specifically, the recommendation system sorts the recommended content in the third recommendation result according to the score of the recommended content, and selects and sorts from the sorted third recommendation result The top a%*k recommended content, the top a%*k recommended content is the recommended content with a higher score, or the recommendation system selects from the sorted A%*k recommended content is selected from the third recommendation result. Similarly, when the recommendation system selects (1-a%)*k recommended content from the second recommendation result, it can also sort the recommended content of the second recommendation result according to the score of the recommended content, and then recommend The top (1-a%)*k recommended content is selected from the results, or the recommendation system selects (1-a%)*k recommended content from the second sorted recommendation result according to a preset principle. Certainly, when the recommendation system selects the recommended content from the third recommendation result and the second recommendation result, it may not sort, or sort according to other parameters, which is not limited in this embodiment of the present invention.

Optionally, in the first way and the second way, after the recommendation system deletes the common recommendation content from the first recommendation result, the recommendation system increases the score of the common recommendation content included in the second recommendation result, correspondingly , when the recommendation system sorts the recommended content in the second recommendation result according to the score, it sorts according to the improved score.

In the third way, firstly, the recommendation system determines the common recommendation content in the first recommendation result and the second recommendation result; secondly, the recommendation system deletes the common recommendation content from the second recommendation result to obtain the fourth recommendation result; then, recommends According to the score of the recommended content, the system ranks the recommended content in the first recommendation result and the fourth recommendation result uniformly; finally, the recommendation system takes the sorted recommended content as the target recommendation result, or the recommendation system uses the preset algorithm from Select part of the recommended content from the sequenced recommended content as the target recommendation result.

In the fourth way, first, the recommendation system determines the common recommendation content in the first recommendation result and the second recommendation result, deletes the common recommendation content from the second recommendation result, and obtains the fourth recommendation result; then, the recommendation system determines the common recommendation content from the first recommendation result Select (1-a%)*k recommended content from the recommended results, and select (1-a%)*k recommended content from the fourth recommended result; finally, the recommendation system selects the recommended content from the first The recommended content selected from the results and the recommended content selected from the fourth recommended result are uniformly sorted, and the uniformly sorted recommended content is used as the target recommendation result.

Among them, the recommendation system selects (1-a%)*k recommended content from the first recommendation result, specifically: the recommendation system sorts the recommended content in the first recommendation result according to the score of the recommended content, and selects the first recommended content from the sorted Selecting the top a%*k recommended content from a recommendation result; or, the recommendation system selects a%*k recommended content from the first sorted recommendation result according to a preset principle. Similarly, when the recommendation system selects (1-a%)*k recommended content from the fourth recommendation result, it can also sort the recommended content in the fourth recommendation result according to the score of the recommended content, and start from the sorted fourth Select (1-a%)*k recommended content from the recommended results, or select (1-a%)*k recommended content from the fourth sorted recommendation result according to a preset principle. Of course, when the recommendation system selects the recommended content from the first recommendation result and the fourth recommendation result, it may not sort, or sort according to other parameters, which is not limited in this embodiment of the present invention.

Optionally, in the third mode and the fourth mode, after the recommendation system deletes the common recommendation content from the second recommendation result, the recommendation system increases the score of the common recommendation content included in the first recommendation result, correspondingly , when the recommendation system sorts the recommended content in the first recommendation result according to the score, it sorts according to the improved score.

Step 205, the recommendation system pushes the target recommendation result to the target user.

In this embodiment, the recommender system generates a cached content library according to the cached content ID and the entire content library by receiving the cached content identifier in the cached content queue sent by the CDN, and subsequently when the recommender system receives the recommendation request message sent by the client , not only obtain the first recommendation result from the entire content library according to the user interest characteristics and the preset recommendation algorithm, but also obtain the second recommendation result from the cached content library according to the user interest characteristics and the recommendation algorithm, and then, according to the preset The fusion algorithm fuses the first recommendation result and the second recommendation result to obtain the target recommendation result, and finally pushes the target recommendation result to the target user. In the method, all recommended content in the second recommendation result is in the cached content queue of the CDN, thereby ensuring that the recommendation results recommended by the recommendation system to the user are in the cached content queue of the CDN as much as possible. When the user requests a recommendation from the CDN When the recommended content is included in the results, the CDN can obtain the content requested by the user from the cache content queue instead of the original server where the content requested by the user is located, reducing the waiting time of the user and reducing the CDN back-to-source Bandwidth usage.

The method of Embodiment 1 will be described in detail below through several specific examples.

Embodiment 2 of the present invention specifically illustrates how the CDN sends the identifier of the cached content in the cached content queue to the recommendation system. FIG. 3 is a schematic structural diagram of the CDN and the recommendation system provided by Embodiment 2 of the invention. As shown in FIG. : Cache content queue receiver, cache content library, recommendation result, all content recommendation popularity library, recommendation popularity sender, push trigger. CDN includes: recommended heat receiver, recommended heat list, push trigger, cached content queue sender, and cached content queue. The identifier of the cached content in the cached content queue of the CDN is sent by the cached content queue sender to the cached content queue receiver of the recommendation system under the control of the push trigger of the CDN, and the recommendation system receives the cached content according to the cached content queue receiver The identifier and the entire content library (not shown in FIG. 3 ) obtain the cached content information, and add the cached content ID and the cached content information to the cached content library, and the recommendation system can establish multiple cached content libraries.

In the embodiment of the present invention, in order to further ensure that the recommendation result recommended by the recommendation system to the user is in the cached content queue of the CDN, the recommendation system is also used to generate the recommendation popularity of all content in the entire content library according to the recommendation result, and the recommendation popularity of the entire content Stored in the recommended popularity library of all content, under the control of the push trigger of the recommendation system, the recommended heat sender sends the data in the recommended heat library of all content to the recommended heat receiver of CDN, and the recommended heat receiver will receive all the data The recommendation popularity of the content is saved in the recommendation popularity list. The embodiment of the present invention introduces the recommendation popularity of the content. The recommendation popularity of the content indicates the possibility that the content will be accessed in the future. The greater the recommendation popularity of the content, the greater the possibility of the content being accessed in the future. frequency.

FIG. 4 is a flow chart of the CDN sending the identifier of cached content to the recommendation system provided by Embodiment 2 of the present invention. As shown in FIG. 4, the method provided by this embodiment may include the following steps:

Step 301, the push trigger of the CDN sends a trigger signal to the cache content queue sender.

In this embodiment, the push trigger of the CDN may send a trigger signal to the cache content queue sender in a periodic manner, so as to trigger the cache content queue sender to send the identifier of the cache content to the recommendation system. There are two types of trigger signals, full trigger signal and incremental trigger signal. When the cache content queue sender receives the full trigger signal, it will put the url or url hash value of p% cache content in front of all cache content queues (according to the recommendation granularity of the recommendation system, the url of the cache content can be the corresponding url or shard url) to the recommendation system. When the cache content queue sender receives the incremental trigger signal, it will send the url of the incremental data of the content of the first P% of the cache content queue relative to the last sent content to the recommendation system. The incremental data is the content of the cache queue The full amount of data is different from the full amount of data in the cache queue content when it was sent last time, and the incremental data is generally much smaller than the full amount of data. For example, assuming that the content stored in the cached content queue at time t1 is "a, b, c, d, e", the CDN sends the urls of these contents to the recommendation system in the form of full data, and the content queue is cached at time t2 The content saved in is "b, c, d, e, f". At this time, the incremental data relative to the time t1 is "a, f", and the CDN sends the url of the incremental data to the recommendation system. After receiving the incremental data, the recommendation system compares the content information "a, b, c, d, e" saved in the cache content library at time t1, and finds that "a" already exists at time t1, then deletes "a" "; find that "f" does not exist at time t1, then add the content of "f" to the cache content library.

In this embodiment, the full trigger signal is generated in two ways: event trigger and periodic trigger. Event trigger refers to the trigger signal generated when the CDN bandwidth is idle; periodic trigger refers to the trigger signal generated every other period T0 (such as 24 hours). These two methods can be used alone or in combination. An example of the combined use of the two is to generate an event trigger signal when the CND bandwidth is idle; if no event trigger signal is generated within the period T0, then generate Periodic trigger signal.

Incremental trigger signals are generated periodically. Generally, the generation period T1 of the incremental trigger signal is shorter than the event trigger time period T0 of the full trigger signal. Full triggering and incremental triggering can also be used in combination. The combination of full triggering and incremental triggering can ensure that the recommendation system obtains the cached content of the accurate cached content queue at a small cost of network traffic. In some cases, such as when the number of content libraries is small or the idle bandwidth of the CDN is abundant, only the full trigger mode can be used.

Step 302, the cache content queue sender judges whether the trigger signal is a full amount trigger signal.

If yes, that is, the trigger signal is a full trigger signal, then execute step 303; if not, that is, the trigger signal is an incremental trigger signal, then execute step 304.

Step 303, the CDN acquires the top p% data of the cached content queue.

The CDN may have multiple cache content queues, and the cache content queues are generally arranged in the order of cache content popularity. In the prior art, the cache content popularity is the cache content access popularity. In the embodiment of the present invention, the cache content Popularity includes access popularity and recommendation popularity of cached content. When the cache content queue at the end of the cache content queue is full, it will be eliminated from the cache content queue. Therefore, in this embodiment, all cached content in the cached content queue cannot be selected to be sent to the recommendation system, so as to avoid content that is about to be eliminated in the recommendation result. Therefore, when the CDN sends the cached content queue data to the recommender system, it only selects the top p% of the cached content queue, so as to ensure that the cached content in the cached queue content recommended by the recommender system is still located in the cached content of the CDN when the user accesses it. In the queue, avoid the prolonged waiting time of users due to the elimination of recommended content. The choice of p% here can define a time interval t (such as 0.5 hours). According to the historical data cached by CDN, calculate the average percentage q% of the content eliminated in the time interval t in the historical data to the cache content queue, then p% =1-q%, or select a percentage based on experience, such as 80% or 90%, or obtain p% according to other parameter adjustment methods (such as using historical data simulation).

The content stored in the cache content queue of CDN can be multimedia objects. Multimedia objects can basically be divided into two categories, one is large multimedia objects such as video and audio, and the other is small multimedia objects such as pictures. . Larger multimedia objects such as video and audio are characterized by the fact that an object is usually divided into multiple fragments and stored in the cache content queue. For large multimedia objects, the client will send a request to the CDN to obtain fragments of related objects; while multimedia objects with small size such as pictures are usually not fragmented, and users access certain pictures through the client. When the multimedia object is displayed, the client will send a request to the CDN to obtain the object. Since the cached content queue of the CDN may store fragmentation information of the object, and the cached content library of the recommendation system stores the information of the object, after the CDN obtains the first p% fragments of the cached content queue, it needs to The fragment is parsed to obtain the information of the object to which the fragment belongs.

In the example shown below, "187.204.219.57--[09/Jul/2014:04:54:58+0000]" GEThttp://sscdn.clarovideo.com/multimediav81/plataforma_vod/ISM/201301/WMP4H01538MTSS_full/WMP4H01538MTSS_full. ism/QualityLevels(1200000)/Fragments(video=3563560000) HTTP/1.1 "200 319213"-""Mozilla/5.0 (WindowsNT 6.2; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/37.0.2008.2 Safari/537.36" "-" is an example of a CDN log generated when a user requests a segment of a large multimedia object such as video and audio, among which "http://sscdn.clarovideo.com/multimediav81/plataforma_vod/ISM/201301/ "WMP4H01538MTSS_full/WMP4H01538MTSS_full.ism" is the url of the multimedia object to which the segment belongs, QualityLevels (1200000) is the code rate information of the multimedia object, (video=3563560000) is the type of the multimedia object and the corresponding timestamp. For multimedia For object-based recommendation systems (such as recommending a certain video or audio), the CDN needs to first parse out the url of the multimedia object to which the p% fragments before the cached content queue belong. When the number of fragments of a multimedia object is greater than a certain threshold (such as 60% of the number of all fragments of the multimedia object), the CDN will consider that the multimedia object is stored in the CDN cache content queue, and then the CDN will send the url of the multimedia object to the recommendation system. For a recommender system with granularity (such as recommending the climax of a certain video), the CDN does not need to parse the urls of the multimedia objects corresponding to the p% fragments before the cached content queue, and directly sends the urls of the requested fragments to recommended system.

In step 304, the CDN acquires incremental data of the top p% data of the cached content queue relative to the content sent last time.

After the CDN determines the top p% data of the cache content queue, it compares the top p% data of the cache content queue with the content sent last time to obtain incremental data.

In step 305, the CDN sends the identifiers of the acquired top p% data or incremental data of the cached content queue to the recommendation system.

It should be noted that the url of the multimedia object generally has a large data volume and needs to be compressed before being sent to the recommendation system. For example, after the url of the multimedia object is hashed, the hash value of the url is sent to the recommendation system.

Step 306 , the recommendation system acquires the information of the cached content according to the received identifier of the cached content and the entire content library, and adds the information of the cached content to the cached content library.

For the specific implementation manner of this step, reference may be made to the relevant description of Embodiment 1, which will not be repeated here.

In this embodiment, the CDN sends the identification of the first P% of the cached content of the cached content queue, or the identification of the incremental data of the first P% of the cached content in the cached content queue, to the recommendation system to ensure that the recommendation system recommends The recommended result for the user is still in the CDN's cached content queue. When the user accesses the recommended result later, the CDN can return the requested content from the cached content queue to the user, thereby reducing the user's waiting time.

Fig. 5 is a flow chart of sending the recommendation heat of the entire content library to the CDN by the recommendation system provided by the third embodiment of the present invention. Please refer to Fig. 3 and Fig. 5, the method provided by this embodiment may include the following steps:

Step 401, the push trigger of the recommendation system sends a trigger signal to the recommendation popularity transmitter.

The recommender system sends recommendation popularity information to the CDN under the control of push triggers, and the recommendation system sends the recommendation popularity information to the CDN in full. The trigger signal is also generated in two ways: event trigger and periodic trigger. Generate a trigger signal. The choice of T can be determined empirically, such as 1 hour. These two trigger methods can be used alone or in combination. An example of the combined use of the two is that the event trigger signal is the main one, and if no event trigger signal is generated within the period T, a periodic trigger signal is generated.

Step 402, the recommendation popularity sender sends the recommendation popularity of all content libraries in the recommendation popularity database to the recommendation popularity receiver of the CDN.

Before sending the recommendation popularity of all content, the recommendation system will also generate a recommendation popularity library according to the recommendation situation of all content in the whole content library. The recommendation popularity library includes all the content in the whole content library within a preset time recommended heat.

In this embodiment, the recommendation system sends the recommendation popularity of all content to the CDN. The purpose of this is to make the CDN consider not only the access popularity of the content, but also the recommendation popularity of the content when eliminating the tail content of the cached content queue. As a result, content that is less popular at present but more likely to be accessed in the future is not easily eliminated from the cache content queue. When a fragment of an object is stored in the cache content queue, the recommendation popularity of the fragment can be the recommendation popularity of the object to which it belongs, or it can be weighted according to the position of the fragment in the object to which it belongs The obtained recommendation popularity, for example, the weight value of the slice at the head of the object is 0.8, the weight value of the middle climax part is 1.0, and the weight value of the tail is 0.7. When calculating the recommendation popularity of a slice, use the slice The recommendation popularity of the object to which it belongs is multiplied by the weight of the shard to obtain the recommendation popularity of the shard.

It should be noted that the recommendation popularity is highly time-sensitive, so the time window W used to calculate the recommendation popularity cannot be too large. One of the methods to determine the size of the time window is to use empirical values, such as 12 hours . The recommendation popularity can be the number of times the content is recommended within the recommendation time window. The recommendation popularity can be represented by one byte. One byte can indicate that the content has been recommended 2^8=1024 times within the recommendation time window at most. The recommendation popularity can also be It is represented by fewer bits, for example, half a byte, and a half byte can at most indicate that the content has been recommended 2^4=16 times within the recommendation time window. In this embodiment, even if data overflow occurs in the recommendation popularity, it will not affect the replacement of the cached content queue by the recommendation popularity, because the content that is recommended more times is likely to be frequently accessed content, and these contents can be relatively stable exists in the cache content queue. Therefore, the main purpose of considering the recommendation popularity when replacing the cached content in the cached content queue is to make the end of the cached content queue less popular as much as possible, but the content that is more likely to be accessed in the future will not be eliminated. As for the content in front of the cache content queue, the present invention does not pay attention to its sorting position, because no matter whether the content in front of the cache content queue is at the front or behind, these contents can be stored in the cache content queue relatively stably.

Step 403 , the recommendation heat receiver of the CDN sends the received recommendation heat of all content to the recommendation heat list.

In this embodiment, the recommendation system sends the recommendation popularity of all content in the entire content library to the CDN, so that when the CDN caches and replaces the cached content queue, it can perform cache replacement according to the recommendation popularity of the cached content, thereby ensuring the current The access popularity is small, but the content that may be accessed in the future is not easy to be eliminated from the cache content queue. In this way, after the recommendation system recommends this part of content to the user, when the user requests this part of content from the CDN, the CDN can retrieve the content from the cache. The content of the request is returned to the user in the queue, thereby reducing the waiting time of the user.

Figure 6 is a schematic diagram of the business process of the method for recommending network content provided by Embodiment 4 of the present invention. As shown in Figure 6, the recommendation system includes a database and a recommendation engine, and the database includes: generated user interest features, the entire content library, and cache Content library, recommendation engine includes core recommendation engine and fusion engine. The core recommendation engine has three inputs: user interest characteristics, the entire content library and the cache content library. When recommending content, the core recommendation engine first obtains the first recommendation result according to the user interest characteristics and the entire content library, and then the core recommendation engine according to The user interest characteristics and the cached content library get the second recommendation result. The first recommendation result and the second recommendation result can be displayed in the form of a list. For example, the first recommendation result is represented by r_1, r_2,..., r_n, and the second recommendation result is represented by rq_1, rq_2,..., rq_m represent. Finally, the fusion engine fuses the first recommendation result and the second recommendation result according to a preset fusion algorithm to obtain a target recommendation result, and outputs the target recommendation result to the client.

FIG. 7 is a flowchart of a method for recommending network content provided in Embodiment 4 of the present invention. Please refer to FIGS. 6 and 7. The method provided in this embodiment may include the following steps:

Step 501, user A views the recommended page.

Step 502, the client sends a recommendation request message to the web server.

The recommendation process is triggered by the user. When the user visits the recommended page or browses a certain object (such as music or video), the browser or player will trigger a recommendation process and send a recommendation request message to the web server.

Step 503, the webpage server sends a query request message for querying user A's recommendation list to the recommendation system.

Step 504, the core recommendation engine of the recommendation system calculates the first recommendation result by using the first recommendation algorithm according to user A's interest characteristics and the entire content library.

Step 505 , the core recommendation engine calculates the second recommendation result by using the second recommendation algorithm according to user A's interest characteristics and cached content library.

The cache content in the cache queue library is generally arranged in the order of updating, that is, the content at the end of the cache content queue will be eliminated from the cache content queue when the cache content queue is full, and the core recommendation engine gets the second according to the cache content queue. When recommending results, you cannot select all cached content in the cached content queue to avoid content that is about to be eliminated in the recommended results. The solution is that the core recommendation engine only selects p% of the cached content in the cached content queue, and the selection method of the p value is the same as that in Embodiment 2, and will not be repeated here.

Step 506, the fusion engine deletes the recommended content common to the second recommendation result from the first recommendation result to obtain the third recommendation result.

Optionally, after step 506, the score of the common recommended content in the second recommendation result may also be increased.

Step 507, the fusion engine uniformly sorts the recommended content in the third recommendation result and the second recommendation result according to the scores of the recommended content.

Step 508, the fusion engine selects K recommended contents from the sorted recommended contents as target recommendation results.

Step 509, the fusion engine updates the entire content recommendation popularity database according to the target recommendation result.

If the recommendation system and CDN work together, and the CDN needs the recommendation popularity sent by the recommendation system to update the cached content queue, the fusion engine will re-determine the recommendation popularity of the recommended content in the target recommendation result according to the target recommendation result, The recommended popularity of the recommended content updates the recommended popularity database of the entire content, otherwise, this step can be omitted.

Step 510, the fusion engine sends the target recommendation result to the web server.

Step 511, the web server forwards the target recommendation result to the client.

Step 512, the client terminal presents the target recommendation result to the user.

FIG. 8 is a flowchart of a method for recommending network content provided in Embodiment 5 of the present invention. As shown in FIG. 8, the method provided in this embodiment may include the following steps:

Step 601, user A views the recommended page.

Step 602, the client sends a recommendation request message to the web server.

Step 603, the webpage server sends a query request message for querying user A's recommendation list to the recommendation system.

Step 604 , the core recommendation engine of the recommendation system calculates the first recommendation result by using the first recommendation algorithm according to the user A's interest characteristics and the entire content library.

Step 605, the core recommendation engine selects the recommended content belonging to the cached content library from the first recommendation result, and takes the selected recommended content as the second recommendation result.

Step 606, the fusion engine deletes the second recommendation result from the first recommendation result to obtain a third recommendation result.

In this embodiment, all the recommended content included in the second recommended result is common recommended content of the first recommended result and the second recommended result. Optionally, after step 606, the score of the recommended content in the second recommendation result may also be increased.

Step 607, the fusion engine selects a%*k recommended content from the third recommended result.

a is a parameter to coordinate the first recommendation result and the second recommendation result: a=0, all recommended content in the target recommendation result comes from the cache content library; a=1, all recommended content in the target recommendation result comes from Sourced from the entire content library. Usually, a can be obtained as a parameter by simulating historical data, or using a machine learning algorithm to calculate an optimal value, or using an online experiment method to obtain an optimal value through user feedback. For a system in which the CDN and the recommendation system work together, the value of a is within the range of (0, 1), that is, a is greater than 0 and less than 1. When determining the specific value of a, for applications that focus on allowing users to have a fast experience, the value of a should be as small as possible without significantly affecting the recommendation effect (such as precision and recall). More recommended content sources cache the content library; for applications that focus on allowing users to have a diverse experience, the value of a should be as large as possible without significantly affecting the speed of user acquisition, that is, to allow more recommended content sources in the entire content library.

Step 608, the fusion engine selects (1-a%)*k recommended content from the second recommendation result.

Step 609 , the fusion engine uniformly sorts the K recommended contents selected from the third recommended result and the second recommended result according to the scores of the recommended contents, and obtains the target recommended result.

Step 610, the fusion engine updates the entire content recommendation popularity database according to the target recommendation result.

Step 611, the fusion engine sends the target recommendation result to the web server.

Step 612, the web server forwards the target recommendation result to the client.

Step 613, the client terminal presents the target recommendation result to the user.

It should be noted that steps 604-606 in this embodiment may be interchanged with steps 504-506 in Embodiment 4, and steps 607-608 in this embodiment may be interchanged with steps 506-507 in Embodiment 4.

Fig. 9 is a schematic structural diagram of a cache content queue and a candidate content queue of a CDN edge server. As shown in Fig. 9, a CDN edge server generally maintains two queues when performing cache replacement: a cache content queue and a candidate content queue. In each embodiment, the cached content queue is used to store content actually cached in the CDN, the candidate content queue is used to store content that may be accessed in the future, and the candidate content stored in the candidate content queue is not cached in the CDN. The CDN edge server has at least one cache content queue, but the number of candidate content queues can be zero, one or more. The update and cache replacement of the cached content queue and the candidate content queue by the CDN edge server are based on the content's access popularity and recommendation popularity. The access popularity of the cached content and candidate content is stored in the access popularity list, and the cached content and candidate content are recommended. The popularity is saved in the recommended popularity list. Of course, the access popularity of cached content and candidate content can also be stored in different access popularity lists, and the recommendation popularity of cached content and candidate content can also be stored in different recommendation popularity lists, which is not limited in the embodiments of the present invention. .

The cached content queue is used to store popular content. The position of the cached content in the cached content queue is sorted according to the popularity. The content's popularity information includes: access popularity, recommendation popularity, calculated based on access popularity and recommendation popularity A combination of comprehensive popularity, visit popularity and recommendation popularity. When the popularity of a certain content in the candidate content queue is greater than a threshold, the CDN edge server will add the content to the cache content queue and delete the content from the candidate content queue. There are many ways to determine the threshold, such as caching The minimum popularity value in the content queue can also be calculated by other methods such as empirical estimation. If the length of the cache content queue reaches a certain threshold (such as the total size of the cache content exceeds 90% of the hard disk capacity), then the least popular part of the cache content queue will be eliminated according to the cache replacement algorithm until the cache content queue The length is less than a certain threshold (such as the total size of the cache content is less than 85% of the hard disk capacity), the eliminated content in the cache content queue is added to the candidate content queue.

The candidate content queue is used to store less popular content, and the candidate content is sorted according to popularity in the candidate content queue. When a certain content is accessed, if the content is not in the cache content queue, save the content in the candidate content queue, calculate its popularity and sort according to the popularity; if the candidate content queue capacity is full, the candidate content queue The content whose end is the same as the total size of the newly added content in the candidate content queue is eliminated from the candidate content queue. There is no fixed ratio between the length of the candidate content queue and the cached content queue, and since the content in the candidate content queue is not stored in the hard disk of the CDN edge server, it does not occupy hard disk storage space. When adding new content to the candidate content queue, if the candidate content queue is full, then eliminate the content with the least popularity in the candidate content queue. The calculation method of the popularity of the candidate content can be the same as that of the cached content, or it can be different.

There are many ways to calculate access popularity, such as the least frequently used page replacement algorithm (Least Frequently Used, referred to as LFU) and the least recently used algorithm (Least Recently Used, referred to as LRU). LRU uses the time difference between the last access time of the content and the current time as the access popularity of the content. At this time, the smaller the time difference between the last access time of the content and the current time, the greater the access popularity of the content. There are also many ways to calculate the recommendation popularity, for example, the number of times a content is recommended within a certain period of time can be used as the access popularity of the content. It should be noted that the recommendation system may use broader information when making recommendations, such as the position and score of a certain video in the popular video rankings. Therefore, the calculation of recommendation popularity may also include such information. In each embodiment of the invention, the recommendation popularity is obtained by the recommendation system according to the recommendation situation, and the recommendation system sends the recommendation popularity of all content to the CDN under the control of the push trigger. There are many ways to calculate the comprehensive popularity. For example, the access popularity and recommendation popularity can be linearly or non-linearly weighted, and the weighted weight can be obtained empirically, or the optimal value can be obtained by machine learning. The combination of visit popularity and recommendation popularity includes: mainly based on visit popularity. In the case of the same visit popularity, the greater the recommendation popularity, the greater the popularity. It can also be based on recommendation popularity. In the case of the same recommendation popularity, the visit The hotter it gets, the hotter it gets.

FIG. 10 is a flow chart of a cache replacement method for network content provided in Embodiment 6 of the present invention. Please refer to FIGS. 9 and 10. The method provided in this embodiment may include the following steps:

Step 701, the CDN edge server obtains the recommendation popularity and access popularity of the cached content in the cached content queue.

Specifically, the recommendation system generates the access popularity of the cached content in the cached content queue according to the historical access conditions of the cached content in the cached content queue. The recommendation system is based on the recommendation popularity of the cached content in the cached content queue sent by the recommendation system. When the CDN sends the recommendation heat, it will send the recommendation heat of all content in the entire content library to the CDN edge server.

In step 702, the CDN edge server performs cache replacement on the cache content queue according to the access popularity and recommendation popularity of the cache content in the cache content queue.

The CDN edge server will periodically detect the size of the remaining space of its own hard disk, for example, every ten minutes. If it is detected that the size of the cache content stored on the hard disk exceeds the first threshold (such as 90% of the hard disk capacity), the CDN edge server will start the cache replacement process until the size of the cache content stored on the hard disk is lower than the second threshold (such as 85% of the hard disk capacity). The cache replacement process of the cache content queue is generally to eliminate the least popular cache content.

In one implementation, if the size of the cached content in the cached content queue is greater than or equal to the first threshold, the CDN edge server determines to eliminate the cached content at the end of the cached content queue with less heat. Usually, the cached content in the cached content queue The cache content is sorted from high to low according to the access popularity. The cache content at the end of the cache content queue has a low access popularity. Some of the cache contents at the end of the cache content queue that need to be eliminated may have the same access popularity. For cached content with the same access popularity, the CDN edge server can further compare the recommendation popularity of the cached content. Specifically, the CDN edge server compares the recommended popularity of the cached content with the same access popularity at the end of the cache content queue, and eliminates the recommended cache content with the same access popularity. If the size is smaller than the second threshold, stop eliminating the cached content queue, and the second threshold is smaller than or equal to the first threshold. For example, the CDN edge server needs to eliminate 10 cached contents at the end of the cached content queue every time, but there are 12 cached contents at the end of the cached content queue with the same access heat, then the CDN edge server compares the recommendations of these 12 cached contents Popularity, out of the 12 cached contents, recommend 10 less popular cached contents.

In another implementation manner, if the size of the cached content in the cached content queue is greater than or equal to the first threshold, the CDN edge server determines to eliminate the cached content at the tail of the cached content queue with less popular access. Then, the CDN edge server calculates the comprehensive heat of the cache content in the tail of the cache content queue according to the access heat and recommendation heat of the cache content at the tail of the cache content queue; the CDN edge server eliminates the comprehensive heat of the cache content in the tail of the queue Small cache content, until the size of the cache content in the cache content queue is smaller than the second threshold, then stop eliminating the cache content queue. For example, the CDN edge server needs to eliminate 10 cached contents at the end of the cached content queue every time, but there are 12 cached contents at the end of the cached content queue with the same access heat, then the CDN edge server can calculate according to the recommendation heat and access heat Based on the comprehensive popularity of the 12 cached contents, 10 cached contents with lower comprehensive popularity are eliminated from the 12 cached contents.

In this embodiment, the CDN edge server obtains the recommendation popularity and access popularity of the cached content in the cached content queue, and when the cached content queue needs to be cached and replaced, the cache is updated according to the cached content's access popularity and recommended popularity in the cached content queue. The content queue does cache replacement. In the method, when performing cache replacement on the cached content queue, not only the access popularity of the cached content, but also the recommendation popularity of the cached content are considered, so that the content with high recommendation popularity is kept in the cache as much as possible, so that the user can Get the content recommended by the recommendation system faster.

Fig. 11 is a flow chart of the cache replacement method for network content provided by Embodiment 7 of the present invention. The difference between Embodiment 6 of this embodiment is that in this embodiment, when the CDN edge server includes both the cached content queue and the candidate content queue , when the CDN edge server caches and replaces the candidate content queue, it also caches and replaces the candidate content queue according to the recommendation popularity of the candidate content. As shown in Figure 11, the method provided in this embodiment may include the following steps:

Step 801, the CDN edge server obtains the recommendation popularity and access popularity of the cached content in the cached content queue.

Step 802, the CDN edge server performs cache replacement on the cache content queue according to the cache content access popularity and recommendation popularity in the cache content queue.

For the specific implementation manner of steps 801 and 802, reference may be made to the description of Embodiment 6, and details are not repeated here.

Step 803, the CDN edge server obtains the recommendation popularity and access popularity of the candidate content in the candidate content queue.

Specifically, the CDN edge server generates the access popularity of the candidate content according to the historical access status of the candidate content in the candidate content queue, and the recommendation system receives the recommendation popularity of the candidate content sent by the recommendation system. generated by the recommendation of the candidate content.

Step 804, the CDN edge server caches and replaces the candidate content queue according to the recommendation popularity and access popularity of the candidate content in the candidate content queue.

In an implementation manner, if the size of the candidate content in the candidate content queue is greater than or equal to the third threshold, the CDN edge server determines to eliminate the candidate content at the tail of the candidate content queue that has less popular access. Usually, the candidate content in the candidate content queue is sorted according to the access popularity from high to low. The access popularity of the candidate content at the tail of the candidate content queue is low, and the access popularity of some candidate content among the candidate content that needs to be eliminated at the tail of the candidate content queue Probably the same. For candidate content with the same access popularity, the CDN edge server can further compare the recommendation popularity of the candidate content. Specifically, the CDN edge server compares the recommendation popularity of the candidate content with the same access popularity in the candidate content queue, and eliminates the candidate content with the lower recommendation popularity among the candidate content with the same access popularity, until the size of the candidate content in the candidate content queue is less than the fourth threshold, stop eliminating the candidate content queue, and the fourth threshold is less than or equal to the third threshold.

In another implementation, if the size of the candidate content in the candidate content queue is greater than or equal to the third threshold, the CDN edge server determines to eliminate the candidate content with less popular access at the tail of the candidate content queue; then, the CDN The edge server calculates the comprehensive popularity of the candidate content in the tail of the candidate content queue according to the access popularity and recommendation popularity of the candidate content in the queue tail of the candidate content queue; finally, compares the comprehensive popularity of the candidate content in the queue tail of the candidate content queue, and eliminates In the tail of the candidate content queue, the candidate content whose comprehensive popularity is small, until the size of the candidate content in the candidate content queue is less than the fourth threshold, then stop eliminating the candidate content queue. The fourth threshold is less than or equal to the third threshold. threshold.

In this embodiment, when cache replacement is performed on the cached content queue and the candidate content queue, not only the access popularity of the cached content and the candidate content is considered, but also the recommendation popularity of the cached content and the candidate content is considered, so as to make the recommendation popularity as high as possible. The content is kept in the cache, so that users can get the content recommended by the recommendation system faster.

When CDN uses recommendation heat for cache replacement, one method is to consider recommendation heat for updating and eliminating the entire cache content queue; another method is to consider recommendation heat only when eliminating the tail of the cache content queue. For example, when eliminating the tail data of the cache content queue with less access popularity, priority is given to eliminating content with the same access popularity but less recommendation popularity. These two methods can also be applied to the cache replacement of the candidate content queue. In the methods of Embodiment 6 and Embodiment 7, the recommendation heat is applied to the process of eliminating the tail of the cache content queue and the candidate content queue. In the following two embodiments, the CDN applies the recommendation heat to the entire cache content Queues are in the process of being updated and eliminated.

FIG. 12 is a flow chart of a cache replacement method for network content provided in Embodiment 8 of the present invention. As shown in FIG. 12 , the method in this embodiment may include the following steps:

In step 901, the CDN edge server receives the content acquisition request sent by the client, and the content acquisition request includes identification information of the content to be accessed.

The CDN’s cache replacement process is triggered by user access. When a user clicks on a recommendation result on a recommended page or a certain content on a terminal application, a cache replacement process is triggered, and the client sends a content acquisition request to the CDN edge server.

It should be noted that the client does not directly send a content acquisition request to the CDN edge server. The client first sends an address resolution request to the DNS server of the CDN. The address resolution request includes the identification information of the content to be accessed. Identify the edge cluster closest to the user. The edge cluster closest to the user includes at least one CDN edge server. When the edge cluster closest to the user includes a CDN edge server, the DNS directly uses the IP address of the CDN edge server. The address is returned to the client, and the client sends a content acquisition request to the CDN edge server. When the edge cluster closest to the user includes multiple CDN edge servers, each CDN edge server has an independent IP address, and only saves a part of the cache content, and the cache content is distributed in a certain way in the cluster, for example, Hash calculation can be performed on the url value of the cached content to obtain the Hash value of the cached content. Different CDN edge servers store cached content corresponding to different Hash value ranges. In this way, each CDN edge server in the edge cluster according to the url of a content Hash value can determine which CDN edge server the content is in. If the DNS knows the range of the hash value of the cached content stored in each CDN edge server in the edge cluster closest to the user, then the DNS server calculates the hash value of the url of the content to be accessed according to the url of the content to be accessed, and according to the The Hash value of the url of the content finds the CDN edge server where the content to be accessed is located from the multiple CDN edge servers, and returns the IP address of the CDN edge server where the content to be accessed is located to the client, and the client sends the request to the CDN where the content to be accessed is located. The edge server sends a content acquisition request. If the DNS knows the range of the hash value of the cached content stored by each CDN edge server in the edge cluster closest to the user, then the DNS server finds the CDN edge server with the least load from the edge cluster closest to the user, and sends the request to the CDN edge server Send a content acquisition request, the CDN edge server with the smallest load finds the CDN edge server where the content to be accessed is located by calculating the Hash value of the url of the content to be accessed, and returns the IP address of the CDN edge server where the content to be accessed is located to the DNS server , the DNS server forwards the IP address of the CDN edge server where the content to be accessed is located to the client, and the client sends content acquisition to the IP address of the CDN edge server where the content to be accessed is based on the IP address of the CDN edge server where the content to be accessed is located ask.

Step 902, the CDN edge server determines whether the content to be accessed is in its cache content queue according to the identification information of the content to be accessed.

If the identification information of the content to be accessed is the url of the content to be accessed, the CDN edge server can judge whether the hash value of the url of the content to be accessed is within its own hash value range, if the hash value of the url of the content to be accessed is within its own hash value If the hash value of the url of the content to be accessed is not within the range of the hash value, it is determined that the content to be accessed is not in the cached content queue. The CDN edge server can also directly determine whether the url of the content to be accessed is in its own cache content queue.

Step 903: If the content to be accessed is in the cached content queue, the CDN edge server returns the content to be accessed to the client.

Step 904: The CDN edge server updates the access popularity of the content to be accessed, calculates the popularity information of the content to be accessed according to the access popularity of the content to be accessed and the recommendation popularity of the content to be accessed, and updates the cached content queue according to the popularity information of the content to be accessed.

Here, the popularity information of the content to be accessed may be the comprehensive popularity of the content to be accessed, or a combination of the popularity of the content to be accessed and the popularity of recommendation. The cached content in the cached content queue is sorted according to the heat information. Here, the cached content queue is updated according to the heat information of the content to be accessed, that is, the order of the cached content in the cached content queue is adjusted according to the heat information of the content to be accessed.

If the cached content in the cached content queue is sorted according to the comprehensive heat, then, calculate the heat information of the content to be accessed according to the heat of the content to be accessed and the recommended heat of the content to be accessed, and update the cached content queue according to the heat information of the content to be accessed, Specifically: the CDN edge server calculates the access popularity and recommendation popularity of the content to be accessed in a linear or non-linear weighted manner to obtain the comprehensive popularity of the content to be accessed, and then, the CDN edge server adjusts the cache content queue according to the comprehensive popularity of the content to be accessed The ordering of the cached content.

If the cache content in the cache content queue is sorted based on the access popularity, when the access popularity of the two cache contents is the same, the recommendation popularity of the two cache contents is compared, and the cache content with the higher recommendation popularity is more popular. The order is before recommending less popular cached content. Then, calculate the popularity information of the content to be accessed according to the popularity of the content to be accessed and the recommendation popularity of the content to be accessed, and update the cache content queue according to the popularity information of the content to be accessed, specifically: the CDN edge server compares the content to be accessed with the cache content queue The access popularity of other cached content in the CDN, if the access popularity of the content to be accessed is not the same as that of other cached content, the CDN edge server directly updates the queue of cached content, if the access popularity of the content to be accessed is the same as that of a certain cached content If the access popularity is the same, the CDN edge server compares the recommendation popularity of the content to be accessed with the recommendation popularity of the cached content with the same access popularity. If the recommendation popularity of the content to be accessed is greater than the recommendation popularity of the cached content with the same access popularity, then The order of the content to be accessed is before the cached content with the same access popularity. If the recommendation popularity of the content to be accessed is lower than the recommendation popularity of the cached content with the same access popularity, the order of the content to be accessed is before the cached content with the same access popularity. After caching the content, if the recommendation popularity of the content to be accessed is equal to the recommendation popularity of the cached content with the same access popularity, the order of the content to be accessed can be before or after the sequence of the cached content with the same access popularity.

If the cache content in the cache content queue is sorted based on recommendation popularity, when the recommendation popularity of two cache contents is the same, the access popularity of the two cache contents is compared, and the cache content with higher access popularity is more popular, when sorting Sequence before accessing less popular cached content. Then, calculate the popularity information of the content to be accessed according to the popularity of the content to be accessed and the recommendation popularity of the content to be accessed, and update the cache content queue according to the popularity information of the content to be accessed, specifically: the CDN edge server compares the content to be accessed with the cache content queue If the recommended popularity of the content to be accessed is not the same as that of other cached content, the CDN edge server will directly update the queue of cached content. If the recommendation popularity is the same, the CDN edge server compares the access popularity of the content to be accessed with the cache content with the same recommendation. If the access popularity of the content to be accessed is greater than that of the cached content with the same recommendation popularity, then The order of the content to be accessed is before the cached content with the same recommendation popularity. After caching the content, if the access popularity of the content to be accessed is equal to the access popularity of the cached content with the same cache popularity, the order of the content to be accessed can be before or after the order of the cached content with the same recommendation popularity.

Step 905: When it is necessary to perform cache replacement on the cached content queue, the CDN edge server eliminates less popular cached content in the cached content queue according to the popularity information of the cached content in the cached content queue.

In this embodiment, when the CDN edge updates the cache content queue, it updates according to the access popularity and recommendation popularity of the cache content. Therefore, the order of all cache content in the cache content queue is sorted according to the access popularity and recommendation popularity. , so that when performing cache replacement on the cache content queue in the future, the cache content with lower popularity at the tail of the cache content queue will be directly eliminated, and it is no longer necessary to calculate the heat information of the cache content at the tail of the queue based on the access heat and recommendation heat of the cache content , sorted according to the calculated popularity information.

In this embodiment, the CDN edge server updates the cache content queue according to the access popularity of the content to be accessed and the recommendation popularity of the content to be accessed, so that when the cache content is cached and replaced, it can ensure that the current access popularity is small, but future access The cached content with high probability is not easy to be eliminated from the cached content queue. When the user accesses the recommendation result later, the CDN can return the requested content from the cached content queue to the user, thereby reducing the waiting time of the user.

Fig. 13 is a flow chart of the cache replacement method for network content provided by Embodiment 9 of the present invention. The difference between this embodiment and Embodiment 8 is that in this embodiment, when the CDN edge server includes both cached content queues and candidate content queues , when the CDN edge server updates the candidate content queue, it also updates the candidate content queue according to the recommendation popularity and access popularity of the candidate content. The method of this embodiment is executed by the CDN edge server, as shown in Figure 13, the method of this embodiment may include the following steps:

Step 1001: Receive a content acquisition request sent by a client, where the content acquisition request includes identification information of the content to be accessed.

Step 1002: Determine whether the content to be accessed is in the cached content queue according to the identification information of the content to be accessed.

If the content to be accessed is in the cached content queue of the CDN edge server, then step 1003 is performed; if the content to be accessed is not in the cached content queue of the CDN edge server, then step 1005 is performed.

Step 1003: Update the popularity of the content to be accessed, determine the popularity information of the content to be accessed according to the popularity of the content to be accessed and the recommendation popularity of the content to be accessed, and update the cached content queue according to the popularity information of the content to be accessed.

After step 1003 is executed, step 1004 is executed.

Step 1004, return the content to be accessed to the client.

For the specific implementation manner of steps 1001-1004, reference may be made to the relevant description of Embodiment 8, which will not be repeated here.

Step 1005. Determine whether the content to be accessed is in the candidate content queue according to the identification information of the content to be accessed.

If the content to be accessed is in the candidate content queue, execute step 1006, and if the content to be accessed is not in the subsequent content queue, execute step 1011. The specific method for the CDN edge server to determine whether the content to be accessed is in the candidate content queue according to the identification information of the content to be accessed is similar to the method for the CDN edge server to determine whether the cached content is in the cached content queue, and will not be repeated here.

Step 1006: Update the access popularity of the content to be accessed, and determine the popularity information of the content to be accessed according to the access popularity and recommendation popularity of the content to be accessed.

Step 1007, judging whether the popularity of the content to be accessed is greater than a preset popularity threshold according to the popularity information of the content to be accessed.

If the popularity of the content to be accessed is greater than the popularity threshold, step 1008 is performed; if the popularity of the content to be accessed is less than or equal to the popularity threshold, step 1010 is performed.

Step 1008, adding the content to be accessed to the cached content queue, and deleting the content to be accessed from the candidate content queue.

Step 1009, returning the IP address of the original server where the content server to be accessed is located to the client.

Step 1010, updating the candidate content queue according to the popularity information of the content to be accessed.

After step 1010 is executed, step 1009 is executed.

Step 1011: Add the content to be accessed to the candidate content queue, update the access popularity of the content to be accessed, and calculate the popularity information of the content to be accessed according to the popularity of the content to be accessed and the recommendation popularity of the content to be accessed.

After step 1011 is executed, step 1010 is executed.

If the information returned by the CDN edge server to the client is the IP address of the original server where the content to be accessed is located, then the client requests the content to be accessed from the original server according to the IP address of the original server, and after obtaining the content to be accessed, Display the content to be accessed to the user.

FIG. 14 is a schematic structural diagram of a recommendation system provided by Embodiment 10 of the present invention. As shown in FIG.

The receiving module 11 is configured to receive the identifier of the cached content in the cached content queue sent by the CDN, obtain the information of the cached content according to the identifier of the cached content and the entire content library, and combine the identifier of the cached content with the cached content Content information is added to the cache content library;

The recommendation module 12 is configured to use a first recommendation algorithm to calculate and obtain a first recommendation result according to the pre-acquired user interest characteristics and the entire content library when the recommendation system receives the recommendation request message sent by the client;

The recommendation module 12 is further configured to obtain a second recommendation result according to the cached content library;

A fusion module 13, configured to fuse the first recommendation result and the second recommendation result according to a preset fusion algorithm to obtain a target recommendation result;

The sending module 14 is configured to push the target recommendation result to the target user.

In an implementation manner, the recommendation module acquires a second recommendation result according to the cached content library, specifically: according to the user interest characteristics and the cached content library, the second recommendation algorithm is used to calculate and obtain the second recommendation result. In another implementation manner, the recommendation module acquires a second recommendation result according to the cached content library, specifically: select recommended content belonging to the cached content library from the first recommendation result, and store the selected recommended content As the second recommendation result.

Optionally, the fusion module 13 is specifically configured to: determine the common recommendation content in the first recommendation result and the second recommendation result; delete the common recommendation content from the first recommendation result to obtain The third recommendation result; according to the score of the recommended content, uniformly sort the recommended content in the second recommended result and the third recommended result; use the sorted recommended content as the target recommended result, or, according to the preset The established algorithm selects part of the recommended content from the sorted recommended content as the target recommendation result.

Optionally, the fusion module 13 is specifically configured to: determine the common recommendation content in the first recommendation result and the second recommendation result; delete the common recommendation content from the first recommendation result to obtain The third recommendation result; select a%*k recommended content from the third recommendation result, where k is the number of recommended content included in the target recommendation result, and a is greater than or equal to 0 and less than or equal to 100; Select (1-a%)*k recommended content from the second recommended result; according to the score of the recommended content, recommend the recommended content selected from the third recommended result and the recommended content selected from the second recommended result The content is sorted uniformly, and the recommended content after the unified sorting is used as the target recommendation result.

Wherein, the fusion module 13 selects a%*k recommended content from the third recommended result, specifically: sorting the recommended content in the third recommended result according to the scores of the recommended content, and sorting the recommended content from the sorted The top a%*k recommended content is selected from the third recommendation result. The fusion module 13 selects (1-a%)*k recommended content from the second recommended result, specifically: sorting the recommended content in the second recommended result according to the scores of the recommended content, from the sorted Select the top (1-a%)*k recommended content in the second recommendation result.

Optionally, after the fusion module 13 deletes the common recommended content from the first recommendation result, it is further configured to: increase the score of the common recommended content included in the second recommendation result.

Optionally, the fusion module 13 is specifically configured to: determine the common recommendation content in the first recommendation result and the second recommendation result; delete the common recommendation content from the second recommendation result to obtain The fourth recommendation result: according to the score of the recommended content, uniformly sort the recommended content in the first recommended result and the fourth recommended result; use the sorted recommended content as the target recommended result, or, according to the preset The established algorithm selects part of the recommended content from the sorted recommended content as the target recommendation result.

Optionally, the fusion module 13 is specifically configured to: determine the common recommendation content in the first recommendation result and the second recommendation result; delete the common recommendation content from the second recommendation result to obtain The fourth recommendation result: select a%*k recommended content from the first recommendation result, where k is the number of recommended content included in the target recommendation result, and a is greater than or equal to 0 and less than or equal to 100; from the Select (1-a%)*k recommended content from the fourth recommended result; according to the score of the recommended content, recommend the recommended content selected from the first recommended result and the recommended content selected from the fourth recommended result The content is sorted uniformly, and the recommended content after the unified sorting is used as the target recommendation result.

Wherein, the fusion module 13 selects a%*k recommended content from the first recommendation result, specifically: sorting the recommended content in the first recommendation result according to the scores of the recommended content, and sorting the recommended content from the sorted Select the top a%*k recommended content in the first recommendation result. The fusion module 13 selects (1-a%)*k recommended content from the fourth recommended result, specifically: sorting the recommended content in the fourth recommended result according to the scores of the recommended content, from the sorted (1-a%)*k recommended contents are selected from the fourth recommended result after that.

Optionally, after the fusion module 13 deletes the common recommended content from the second recommendation result, it is further configured to: increase the score of the common recommended content included in the first recommendation result.

Optionally, the recommendation system further includes: a recommendation popularity generating module, configured to generate a recommendation popularity library according to the recommendation situation of all contents in the entire content library, and the recommendation popularity library includes the recommendation popularity library in the entire content library The recommended popularity of all content within a preset time; the sending module 14 is also configured to send all the content in the recommended popularity library to the CDN. Correspondingly, the fusion module 13 fuses the first recommendation result and the second recommendation result according to a preset fusion algorithm, and after obtaining the target recommendation result, the recommendation popularity generation module is further used to: according to the The target recommendation result updates the recommendation popularity library.

Optionally, in this embodiment, the cache content sent by the CDN is the content of the top P% of the cache content queue, or the content of the top P% of the cache content queue relative to the content sent last time Incremental data, where P is greater than 0 and less than 100.

The recommendation system of this embodiment can be used to implement the methods of Embodiment 1 to Embodiment 5, and the specific implementation manner and technical effect are similar, and will not be repeated here.

FIG. 15 is a schematic structural diagram of a CDN edge server provided by Embodiment 11 of the present invention. As shown in FIG. 15 , the CDN edge server provided by this embodiment includes: an acquisition module 21 and a cache replacement module 22 .

An acquisition module 21, configured to acquire recommendation popularity and access popularity of cached content in the cached content queue;

The cache replacement module 22 is configured to perform cache replacement on the cache content queue according to the access popularity and recommendation popularity of the cache content in the cache content queue.

Optionally, the cache replacement module 22 is specifically configured to: if the size of the cached content in the cached content queue is greater than or equal to a first threshold, determine a cache with less popularity for accessing the tail of the cached content queue The content is eliminated; compare the recommendation heat of the cache content with the same access heat at the tail of the cache content queue, and eliminate the less recommended cache content in the cache content with the same access heat, until the cache content in the cache content queue If the size of the cached content is smaller than a second threshold, the elimination of the cached content queue is stopped, and the second threshold is smaller than or equal to the first threshold.

Optionally, the cache replacement module 22 is specifically configured to: if the size of the cached content in the cached content queue is greater than or equal to a first threshold, determine that the access to the tail of the cached content queue is less popular The cache content is eliminated; according to the access heat and recommendation heat of the cache content at the tail of the cache content queue, calculate the comprehensive heat of the cache content in the tail of the cache content queue; eliminate the cache content at the tail of the cache content queue Combining less popular cached content until the size of the cached content in the cached content queue is less than a second threshold, stop eliminating the cached content queue, the second threshold being less than or equal to the first threshold.

In this embodiment, the acquisition module 21 is specifically configured to: generate the access heat of the cached content in the cached content queue according to the historical access conditions of the cached content in the cached content queue; receive the cached content sent by the recommendation system The recommendation popularity of the cached content in the content queue, the recommendation popularity of the cached content in the cached content queue is generated by the recommendation system according to the recommendation of the cached content in the cached content queue.

The CDN edge server provided in this embodiment can be used to implement the method in Embodiment 6, and the specific implementation manner and technical effect are similar, and will not be repeated here.

Embodiment 12 of the present invention provides a CDN edge server. The structure of the CDN edge server provided in this embodiment is the same as that shown in FIG. 15. Please refer to FIG. 15. In this embodiment, the acquisition module 21 also It is used to: obtain the recommendation popularity and access popularity of the candidate content in the candidate content queue; the cache replacement module 22 is also used to: according to the recommendation popularity and access popularity of the candidate content in the candidate content queue Queue for cache replacement.

Optionally, the cache replacement module 22 is specifically configured to: if the size of the candidate content in the candidate content queue is greater than or equal to a third threshold, determine that the access to the tail of the candidate content queue is less popular The candidate content is eliminated; comparing the recommendation heat of the candidate content with the same access heat in the candidate content queue, eliminating the candidate content with the lower recommendation heat among the candidate content with the same access heat, until the candidate content in the candidate content queue If the size of the candidate content is smaller than a fourth threshold, the elimination of the candidate content queue is stopped, and the fourth threshold is smaller than or equal to the third threshold.

Optionally, the cache replacement module 22 is specifically configured to: if the size of the candidate content in the candidate content queue is greater than or equal to a third threshold, determine that the access to the tail of the candidate content queue is less popular Candidate content is eliminated; According to the access popularity and recommendation popularity of the candidate content in the queue tail of the candidate content queue, the comprehensive popularity of the candidate content in the queue tail of the candidate content queue is calculated; the candidate content queue is eliminated. Candidate content whose comprehensive popularity is less, stop eliminating the candidate content queue until the size of the candidate content in the candidate content queue is less than the fourth threshold, and the fourth threshold is less than or equal to the first Three thresholds.

In this embodiment, the acquisition module 21 is specifically configured to: generate the access popularity of the candidate content in the candidate content queue according to the historical access conditions of the candidate content in the candidate content queue; receive the candidate content sent by the recommendation system. The recommendation popularity of the content, the recommendation popularity of the candidate content is generated by the recommendation system according to the recommendation situation of the candidate content.

The CDN edge server provided in this embodiment can be used to implement the method in Embodiment 7, and the specific implementation manner and technical effect are similar, and will not be repeated here.

FIG. 16 is a schematic structural diagram of a CDN edge server provided by Embodiment 13 of the present invention. As shown in FIG. 16 , the CDN edge server provided by this embodiment includes: a receiving module 31, a processing module 32, an update module 33 and a cache Replace module 34.

A receiving module 31, configured to receive a content acquisition request sent by the client, where the content acquisition request includes identification information of the content to be accessed;

The processing module 32 is configured to determine whether the content to be accessed is in its own cache content queue according to the identification information of the content to be accessed, and if the content to be accessed is in the cache content queue, send the return the content to be accessed;

An update module 33, configured to update the access popularity of the content to be accessed, and calculate the popularity information of the content to be accessed according to the access popularity and recommendation popularity of the content to be accessed, and update the content according to the popularity information of the content to be accessed The above cache content queue;

The cache replacement module 34 is configured to, when it is necessary to perform cache replacement on the cache content queue, eliminate less popular cache content in the cache content queue according to the popularity information of the cache content in the cache content queue.

If the identification information of the content to be accessed is not in the cached content queue, the processing module 32 is further configured to: determine whether the content to be accessed is a candidate for the CDN edge server according to the identification information of the content to be accessed In the content queue; if the content to be accessed is in the candidate content queue, update the access popularity of the content to be accessed, and determine the popularity information of the content to be accessed according to the access popularity and recommendation popularity of the content to be accessed ; Judging whether the popularity of the content to be accessed is greater than a preset popularity threshold according to the popularity information of the content to be accessed; if the popularity of the content to be accessed is greater than the popularity threshold, adding the content to be accessed to the and delete the to-be-accessed content from the candidate content queue; and return the IP address of the original server where the to-be-accessed content server is located to the client.

If the content to be accessed is not in the candidate content queue, the processing module 32 is further configured to: add the content to be accessed to the candidate content queue; update the access popularity of the content to be accessed, according to the Determine the popularity information of the content to be accessed according to the popularity of the content to be accessed and the recommendation popularity of the content to be accessed, update the queue of candidate content according to the popularity information of the content to be accessed; return the content to be accessed to the client Access the IP address of the origin server where the content server is located.

If the popularity of the content to be accessed is less than or equal to the popularity threshold, the processing module 32 is further configured to: update the candidate content queue according to the popularity information of the content to be accessed; return the IP address of the original server where the content server to be accessed is located.

Optionally, the cache replacement module 34 is further configured to: when it is necessary to perform cache replacement on the candidate content queue, according to the popularity information of the candidate content in the candidate content queue, eliminate the candidates with less popularity in the candidate content queue candidate content.

The CDN edge server provided in this embodiment can be used to implement the methods of Embodiment 8 and Embodiment 9, and the specific implementation manner and technical effect are similar, and will not be repeated here.

FIG. 17 is a schematic structural diagram of a CDN edge server provided in Embodiment 14 of the present invention. As shown in FIG. 17 , the CDN edge server 400 provided in this embodiment includes: a processor 41, a memory 42, a communication interface 43, and a system bus 44. The memory 42 and the communication interface 43 are connected and communicate with the processor 41 through the system bus 44; the memory 42 is used to store computer execution instructions; the communication interface 43 is used to communicate with other devices , the processor 41 is configured to run the computer to execute instructions and perform the following method:

Obtain the recommendation popularity and access popularity of the cached content in the cached content queue;

Perform cache replacement on the cache content queue according to the access popularity and recommendation popularity of the cache content in the cache content queue.

Optionally, the processor 41 performs cache replacement on the cache content queue according to the cache content access and recommendation popularity in the cache content queue, specifically: if the size of the cache content in the cache content queue is greater than or equal to the first threshold, it is determined to eliminate the cache content with less access heat at the tail of the cache content queue; then, compare the recommendation heat of the cache content with the same access heat at the tail of the cache content queue, Eliminate cached content with less recommended popularity among the cached content with the same access popularity, until the size of the cached content in the cached content queue is smaller than a second threshold, then stop eliminating the cached content queue, the first The second threshold is less than or equal to the first threshold.

Optionally, the processor 41 performs cache replacement on the cache content queue according to the cache content access and recommendation popularity in the cache content queue, specifically: if the size of the cache content in the cache content queue is If it is greater than or equal to the first threshold, it is determined to eliminate the cached content with less popular access at the tail of the cache content queue; then, according to the access popularity and recommendation popularity of the cached content at the tail of the cache content queue, Calculate the comprehensive heat of the cache content in the queue tail of the cache content queue; finally, eliminate the cache content with less comprehensive heat in the queue tail of the cache content queue until the size of the cache content in the cache content queue is less than the first If the second threshold is lower than or equal to the first threshold, the elimination of the cached content queue is stopped.

In this embodiment, the processor 41 acquires the recommendation popularity and access popularity of the cached content in the cached content queue, specifically: generating the cached content in the cached content queue according to the historical access status of the cached content in the cached content queue The access heat of cached content; receiving the recommended heat of cached content in the cached content queue sent by the recommender system, the recommended heat of cached content in the cached content queue is the recommendation system according to the cached content in the cached content queue Content recommendations are generated.

The processor 41 is further configured to: acquire recommendation popularity and access popularity of candidate content in the candidate content queue; perform cache replacement on the candidate content queue according to the recommendation popularity and access popularity of candidate content in the candidate content queue.

Optionally, the processor 41 caches and replaces the candidate content queue according to the recommendation popularity and access popularity of the candidate content in the candidate content queue, specifically: if the size of the candidate content in the candidate content queue is greater than or equal to the third threshold, then it is determined that the candidate content with less access popularity at the tail of the candidate content queue is eliminated; then, compare the recommendation popularity of the candidate content with the same access popularity in the candidate content queue, and eliminate Recommend less popular candidate content among the candidate contents with the same access popularity, and stop eliminating the candidate content queue until the size of the candidate content in the candidate content queue is smaller than a fourth threshold, and the fourth The threshold is less than or equal to the third threshold.

Optionally, the processor 41 caches and replaces the candidate content queue according to the recommendation popularity and access popularity of the candidate content in the candidate content queue, specifically: if the size of the candidate content in the candidate content queue is greater than or equal to the third threshold, it is determined to eliminate the candidate content with less popular access at the tail of the candidate content queue; then, according to the access popularity and recommendation popularity of the candidate content at the tail of the candidate content queue, Calculate the comprehensive heat of the candidate content in the queue tail of the candidate content queue; finally, eliminate the candidate content with less comprehensive heat of the candidate content in the queue tail of the candidate content queue until the candidate content in the candidate content queue If the size is smaller than a fourth threshold, stop eliminating the candidate content queue, and the fourth threshold is smaller than or equal to the third threshold.

In this embodiment, the processor 41 obtains the recommendation popularity and access popularity of the candidate content in the candidate content queue, specifically: generating the candidate content in the candidate content queue according to the historical access status of the candidate content in the candidate content queue The access popularity of the candidate content: receiving the recommendation popularity of the candidate content sent by the recommendation system, the recommendation popularity of the candidate content is generated by the recommendation system according to the recommendation situation of the candidate content.

The CDN edge server of this embodiment can be used to execute the methods of Embodiment 6 and Embodiment 7, and the specific implementation manner and technical effect are similar, and will not be repeated here.

FIG. 18 is a schematic structural diagram of a CDN edge server provided in Embodiment 15 of the present invention. As shown in FIG. 18 , the CDN edge server 500 provided in this embodiment includes: a processor 51, a memory 52, a communication interface 53, and a system bus 54. The memory 52 and the communication interface 53 are connected and communicate with the processor 51 through the system bus 54; the memory 52 is used to store computer execution instructions; the communication interface 53 is used to communicate with other devices , the processor 51 is configured to run the computer to execute instructions and execute the method described below:

receiving a content acquisition request sent by the client, where the content acquisition request includes identification information of the content to be accessed;

Determine whether the content to be accessed is in its own cache content queue according to the identification information of the content to be accessed;

If the content to be accessed is in the cached content queue, return the content to be accessed to the client;

updating the access popularity of the content to be accessed, and calculating the popularity information of the content to be accessed according to the access popularity and recommendation popularity of the content to be accessed, and updating the cached content queue according to the popularity information of the content to be accessed;

When it is necessary to perform cache replacement on the cache content queue, the cache content with less popularity in the cache content queue is eliminated according to the popularity information of the cache content in the cache content queue.

If the identification information of the content to be accessed is not in the cached content queue, the processor 51 is further configured to: determine whether the content to be accessed is a candidate for the CDN edge server according to the identification information of the content to be accessed In the content queue; if the content to be accessed is in the candidate content queue, update the access popularity of the content to be accessed, and determine the popularity information of the content to be accessed according to the access popularity and recommendation popularity of the content to be accessed ; Judging whether the popularity of the content to be accessed is greater than a preset popularity threshold according to the popularity information of the content to be accessed; if the popularity of the content to be accessed is greater than the popularity threshold, adding the content to be accessed to the and delete the to-be-accessed content from the candidate content queue; and return the IP address of the original server where the to-be-accessed content server is located to the client.

If the content to be accessed is not in the candidate content queue, the processor 51 is further configured to: add the content to be accessed to the candidate content queue; update the access popularity of the content to be accessed, according to the Determine the popularity information of the content to be accessed according to the popularity of the content to be accessed and the recommendation popularity of the content to be accessed, update the queue of candidate content according to the popularity information of the content to be accessed; return the content to be accessed to the client Access the IP address of the origin server where the content server is located.

If the popularity of the content to be accessed is less than or equal to the popularity threshold, the processor 51 is further configured to: update the candidate content queue according to the popularity information of the content to be accessed; return the IP address of the original server where the content server to be accessed is located.

The processor 51 is further configured to: when it is necessary to cache and replace the candidate content queue, eliminate less popular candidate content in the candidate content queue according to popularity information of the candidate content in the candidate content queue.

FIG. 19 is a schematic structural diagram of a recommendation system provided by Embodiment 16 of the present invention. As shown in FIG. 19 , the recommendation system 600 provided by this embodiment includes: a processor 61, a memory 62, a communication interface 63, and a system bus 64. The memory 62 and the communication interface 63 are connected and communicate with the processor 61 through the system bus 64; the memory 62 is used to store computer execution instructions; the communication interface 63 is used to communicate with other devices, so The processor 61 is configured to run the computer to execute instructions and perform the following methods:

Receive the identifier of the cached content in the cached content queue sent by the CDN, obtain the information of the cached content according to the identifier of the cached content and the entire content library, and add the identifier of the cached content and the information of the cached content to the cache content library;

When the recommendation system receives the recommendation request message sent by the client, according to the pre-acquired user interest characteristics and the entire content library, the first recommendation algorithm is used to calculate and obtain the first recommendation result;

Obtaining a second recommendation result according to the cached content library;

merging the first recommendation result and the second recommendation result according to a preset fusion algorithm to obtain a target recommendation result;

Pushing the target recommendation result to the target user.

Optionally, the processor 61 obtains the second recommendation result according to the cached content library, specifically: according to the user interest characteristics and the cached content library, calculate the second recommendation result by using a second recommendation algorithm .

Optionally, the processor 61 acquires the second recommendation result according to the cached content library, specifically: selecting recommended content belonging to the cached content library from the first recommendation result, and using the selected recommended content as the selected Describe the results of the second recommendation.

Optionally, the processor 61 fuses the first recommendation result and the second recommendation result according to a preset fusion algorithm to obtain a target recommendation result, specifically: determining the first recommendation result and the common recommendation content in the second recommendation result; delete the common recommendation content from the first recommendation result to obtain a third recommendation result; The recommended content in the recommendation results is uniformly sorted; the sorted recommended content is used as the target recommendation result, or part of the recommended content is selected from the sorted recommended content according to a preset algorithm as the target recommendation result.

Optionally, the processor 61 fuses the first recommendation result and the second recommendation result according to a preset fusion algorithm to obtain a target recommendation result, specifically: determining the first recommendation result and the Common recommended content in the second recommended result; delete the common recommended content from the first recommended result to obtain a third recommended result; select a%*k recommended content from the third recommended result, wherein , k is the number of recommended content included in the target recommendation result, a is greater than or equal to 0 and less than or equal to 100; select (1-a%)*k recommended content from the second recommended result; according to the recommended content Scoring, performing unified ranking on the recommended content selected from the third recommended result and the recommended content selected from the second recommended result, and using the unified ranked recommended content as the target recommended result.

Wherein, the processor 61 selects a%*k recommended content from the third recommendation result, specifically: sorting the recommended content in the third recommendation result according to the scores of the recommended content, and sorting The top a%*k recommended content is selected from the third recommendation result. The processor 61 selects (1-a%)*k recommended content from the second recommendation result, specifically: sorting the recommended content in the second recommendation result according to the scores of the recommended content, and sorting Select the top (1-a%)*k recommended content in the second recommendation result.

Optionally, after the processor 61 deletes the common recommended content from the first recommendation result, it is further configured to: increase the score of the common recommended content included in the second recommendation result.

Optionally, the processor 61 fuses the first recommendation result and the second recommendation result according to a preset fusion algorithm to obtain a target recommendation result, specifically: determining the first recommendation result and the common recommendation content in the second recommendation result; delete the common recommendation content from the second recommendation result to obtain a fourth recommendation result; The recommended content in the recommendation results is uniformly sorted; the sorted recommended content is used as the target recommendation result, or part of the recommended content is selected from the sorted recommended content according to a preset algorithm as the target recommendation result.

Optionally, the processor 61 fuses the first recommendation result and the second recommendation result according to a preset fusion algorithm to obtain a target recommendation result, specifically: determining the first recommendation result and the Common recommendation content in the second recommendation result; delete the common recommendation content from the second recommendation result to obtain the fourth recommendation result; select a%*k recommendation content from the first recommendation result, wherein , k is the number of recommended content included in the target recommendation result, a is greater than or equal to 0 and less than or equal to 100; select (1-a%)*k recommended content from the fourth recommended result; according to the recommended content Scoring, performing unified ranking on the recommended content selected from the first recommended result and the recommended content selected from the fourth recommended result, and using the unified ranked recommended content as the target recommended result.

Wherein, the processor 61 selects a%*k recommended content from the first recommendation result, specifically: sorting the recommended content in the first recommendation result according to the scores of the recommended content, and sorting Select the top a%*k recommended content in the first recommendation result. The processor 61 selects (1-a%)*k recommended content from the fourth recommendation result, specifically: sorting the recommended content in the fourth recommendation result according to the scores of the recommended content, and sorting (1-a%)*k recommended contents are selected from the fourth recommended result after that.

Optionally, after the processor 61 deletes the common recommended content from the second recommendation result, it is further configured to: increase the score of the common recommended content included in the first recommendation result.

Optionally, the processor 61 is further configured to: generate a recommendation popularity library according to the recommendation situation of all the contents in the entire content library, and the recommendation popularity library includes all the contents in the entire content library in the preset The recommended popularity within a certain period of time; sending all the content in the recommended popularity library to the CDN. Correspondingly, the processor 61 fuses the first recommendation result and the second recommendation result according to a preset fusion algorithm, and after obtaining a target recommendation result, is further configured to: update the Recommended heat library.

In this embodiment, the cache content sent by the CDN is the content of the top P% of the cache content queue, or the incremental data of the top P% content of the cache content queue relative to the content sent last time , where P is greater than 0 and less than 100.

The recommendation system provided in this embodiment can be used to implement the methods of Embodiment 1 to Embodiment 5, and the specific implementation manner and technical effect are similar, and will not be repeated here.

Those of ordinary skill in the art can understand that all or part of the steps for implementing the above method embodiments can be completed by program instructions and related hardware. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above-mentioned method embodiments; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk and other various media that can store program codes.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: It is still possible to modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the various embodiments of the present invention. scope.

Claims (54)

1. A method for recommending network content, comprising: The recommendation system receives the identifier of the cached content in the cached content queue sent by the content distribution network CDN, obtains the information of the cached content according to the identifier of the cached content and the entire content library, and combines the identifier of the cached content with the cached content The information added to the cache content library; When the recommendation system receives the recommendation request message sent by the client, the recommendation system uses the first recommendation algorithm to calculate the first recommendation result according to the pre-acquired user interest characteristics and the entire content library; The recommendation system acquires a second recommendation result according to the cached content library; The recommendation system fuses the first recommendation result and the second recommendation result according to a preset fusion algorithm to obtain a target recommendation result; The recommendation system pushes the target recommendation result to the target user. 2. The method according to claim 1, wherein the recommendation system obtains a second recommendation result according to the cached content library, comprising: The recommendation system calculates the second recommendation result by using a second recommendation algorithm according to the user interest characteristics and the cached content library. 3. The method according to claim 1, wherein the recommendation system obtains a second recommendation result according to the cached content library, comprising: The recommendation system selects recommended content belonging to the cached content library from the first recommendation result, and takes the selected recommended content as the second recommendation result. 4. The method according to any one of claims 1-3, wherein the recommendation system fuses the first recommendation result and the second recommendation result according to a preset fusion algorithm to obtain the target Recommended results, including: The recommendation system determines common recommendation content in the first recommendation result and the second recommendation result; The recommendation system deletes the common recommendation content from the first recommendation result to obtain a third recommendation result; The recommendation system uniformly sorts the recommended content in the second recommendation result and the third recommendation result according to the score of the recommended content; The recommendation system takes the sorted recommended content as the target recommendation result, or the recommendation system selects part of the recommended content from the sorted recommended content according to a preset algorithm as the target recommendation result. 5. The method according to any one of claims 1-3, wherein the recommendation system fuses the first recommendation result and the second recommendation result according to a preset fusion algorithm to obtain the target Recommended results, including: The recommendation system determines common recommendation content in the first recommendation result and the second recommendation result; The recommendation system deletes the common recommendation content from the first recommendation result to obtain a third recommendation result; The recommendation system selects a%*k recommended content from the third recommendation result, where k is the number of recommended content included in the target recommendation result, and a is greater than or equal to 0 and less than or equal to 100; The recommendation system selects (1-a%)*k recommended content from the second recommendation result; The recommendation system uniformly sorts the recommended content selected from the third recommendation result and the recommended content selected from the second recommendation result according to the scores of the recommended content, and uses the unified sorted recommended content as the Target recommendation results. 6. The method according to claim 5, wherein the recommendation system selects a%*k recommended content from the third recommendation result, including: The recommendation system sorts the recommended content in the third recommendation result according to the score of the recommended content, and selects the top a%*k recommended content from the sorted third recommendation result; The recommendation system selects (1-a%)*k recommended content from the second recommendation result, including: The recommendation system sorts the recommended content in the second recommendation result according to the score of the recommended content, and selects the top (1-a%)*k recommended content from the sorted second recommendation result . 7. The method according to any one of claims 4-6, wherein after the recommendation system deletes the common recommendation content from the first recommendation result, the method further comprises: The recommendation system increases the score of the common recommended content included in the second recommendation result. 8. The method according to any one of claims 1-3, wherein the recommendation system fuses the first recommendation result and the second recommendation result according to a preset fusion algorithm to obtain the target Recommended results, including: The recommendation system determines common recommendation content in the first recommendation result and the second recommendation result; The recommendation system deletes the common recommendation content from the second recommendation result to obtain a fourth recommendation result; The recommendation system uniformly sorts the recommended content in the first recommendation result and the fourth recommendation result according to the score of the recommended content; The recommendation system takes the sorted recommended content as the target recommendation result, or, the recommendation system selects part of the recommended content from the sorted recommended content according to a preset algorithm as the target recommendation result. 9. The method according to any one of claims 1-3, wherein the recommendation system fuses the first recommendation result and the second recommendation result according to a preset fusion algorithm to obtain the target Recommended results, including: The recommendation system determines common recommendation content in the first recommendation result and the second recommendation result; The recommendation system deletes the common recommendation content from the second recommendation result to obtain a fourth recommendation result; The recommendation system selects a%*k recommended content from the first recommendation result, where k is the number of recommended content included in the target recommendation result, and a is greater than or equal to 0 and less than or equal to 100; The recommendation system selects (1-a%)*k recommended content from the fourth recommendation result; The recommendation system uniformly sorts the recommended content selected from the first recommendation result and the recommended content selected from the fourth recommendation result according to the scores of the recommended content, and uses the unified sorted recommended content as the recommended content. The target recommendation results. 10. The method according to claim 9, wherein the recommendation system selects a%*k recommended content from the first recommendation result, including: The recommendation system sorts the recommended content in the first recommendation result according to the score of the recommended content, and selects the top a%*k recommended content from the sorted first recommendation result; The recommendation system selects (1-a%)*k recommended content from the fourth recommendation result, including: The recommendation system sorts the recommended content in the fourth recommendation result according to the score of the recommended content, and selects (1-a%)*k recommended content from the sorted fourth recommendation result. 11. The method according to any one of claims 7-10, wherein after the recommendation system deletes the common recommendation content from the second recommendation result, the method further comprises: The recommendation system increases the score of the common recommended content included in the first recommendation result. 12. The method according to any one of claims 1-3, further comprising: The recommendation system generates a recommendation popularity library according to the recommendation situation of all the contents in the entire content library, and the recommendation popularity library includes the recommendation popularity of all the contents in the entire content library within a preset time; The recommendation system sends all the content in the recommended popularity library to the CDN. 13. The method according to claim 12, wherein the recommendation system fuses the first recommendation result and the second recommendation result according to a preset fusion algorithm, and after obtaining the target recommendation result, the Methods also include: The recommendation system updates the recommendation popularity library according to the target recommendation result. 14. The method according to any one of claims 1-13, wherein the cache content sent by the CDN is the content of the top P% of the cache content queue, or is the content of the cache content queue Incremental data of the content of the top P% relative to the content sent last time, where P is greater than 0 and less than 100. 15. A cache replacement method for network content, comprising: The content distribution network CDN edge server obtains the recommendation popularity and access popularity of the cached content in the cached content queue; The CDN edge server performs cache replacement on the cache content queue according to the access popularity and recommendation popularity of the cache content in the cache content queue. 16. The method according to claim 15, wherein the CDN edge server caches and replaces the cached content queue according to the access popularity and recommendation popularity of the cached content in the cached content queue, comprising: If the size of the cached content in the cached content queue is greater than or equal to the first threshold, the CDN edge server determines to eliminate the cached content at the end of the cached content queue with less popular access; The CDN edge server compares the recommended popularity of the cached content with the same access popularity at the tail of the cache content queue, and eliminates the cache content with the lower recommended popularity among the cache contents with the same access popularity, until the cache content queue If the size of the cached content in the queue is smaller than a second threshold, the elimination of the cached content queue is stopped, and the second threshold is smaller than or equal to the first threshold. 17. The method according to claim 15, wherein the CDN edge server caches and replaces the cached content queue according to the access popularity and recommendation popularity of the cached content in the cached content queue, comprising: If the size of the cached content in the cached content queue is greater than or equal to the first threshold, the CDN edge server determines to eliminate the cached content at the queue tail of the cached content queue with less popular access; The CDN edge server calculates the comprehensive popularity of the cache content at the tail of the cache content queue according to the access popularity and recommendation popularity of the cache content at the tail of the cache content queue; The CDN edge server eliminates cached content with low overall popularity in the tail of the cached content queue until the size of the cached content in the cached content queue is smaller than a second threshold, then stops eliminating the cached content queue , the second threshold is less than or equal to the first threshold. 18. The method according to claim 15, wherein the CDN edge server acquires recommendation popularity and access popularity of cached content in the cached content queue, comprising: The CDN edge server generates the access heat of the cached content in the cached content queue according to the historical access conditions of the cached content in the cached content queue; The CDN edge server receives the recommendation heat of the cache content in the cache content queue sent by the recommendation system, and the recommendation heat of the cache content in the cache content queue is the recommendation system according to the cache content in the cache content queue generated by the recommendations. 19. The method according to any one of claims 15-18, further comprising: The CDN edge server acquires recommendation popularity and access popularity of candidate content in the candidate content queue; The CDN edge server performs cache replacement on the candidate content queue according to the recommendation popularity and access popularity of the candidate content in the candidate content queue. 20. The method according to claim 19, wherein the CDN edge server caches and replaces the candidate content queue according to the recommendation popularity and access popularity of the candidate content in the candidate content queue, comprising: If the size of the candidate content in the candidate content queue is greater than or equal to a third threshold, the CDN edge server determines to eliminate the candidate content at the tail of the candidate content queue that has less popular access; The CDN edge server compares the recommendation popularity of the candidate content with the same access popularity in the candidate content queue, and eliminates the candidate content with the lower recommendation popularity among the candidate content with the same access popularity, until the candidate content in the candidate content queue If the size of the candidate content is smaller than a fourth threshold, the elimination of the candidate content queue is stopped, and the fourth threshold is smaller than or equal to the third threshold. 21. The method according to claim 19, wherein the CDN edge server caches and replaces the candidate content queue according to the recommendation popularity and access popularity of the candidate content in the candidate content queue, comprising: If the size of the candidate content in the candidate content queue is greater than or equal to a third threshold, the CDN edge server determines to eliminate the candidate content at the tail of the candidate content queue that has less popular access; The CDN edge server calculates the comprehensive popularity of the candidate content in the queue tail of the candidate content queue according to the access popularity and recommendation popularity of the candidate content in the queue tail of the candidate content queue; The CDN edge server eliminates the candidate content with lower comprehensive popularity of the candidate content in the tail of the candidate content queue until the size of the candidate content in the candidate content queue is smaller than the fourth threshold, then stops processing the candidate content The queue is eliminated, and the fourth threshold is less than or equal to the third threshold. 22. The method according to claim 19, wherein the CDN edge server acquires recommendation popularity and access popularity of candidate content in the candidate content queue, comprising: The CDN edge server generates the access heat of the candidate content in the candidate content queue according to the historical access conditions of the candidate content in the candidate content queue; The recommendation system receives the recommendation popularity of the candidate content sent by the recommendation system, and the recommendation popularity of the candidate content is generated by the recommendation system according to the recommendation situation of the candidate content. 23. A cache replacement method for network content, comprising: The edge server of the content distribution network CDN receives the content acquisition request sent by the client, and the content acquisition request includes identification information of the content to be accessed; The CDN edge server determines whether the content to be accessed is in its cache content queue according to the identification information of the content to be accessed; If the content to be accessed is in the cached content queue, the CDN edge server returns the content to be accessed to the client; The CDN edge server updates the access popularity of the content to be accessed, and calculates the popularity information of the content to be accessed according to the access popularity and recommendation popularity of the content to be accessed, and updates the Cache content queue; When it is necessary to perform cache replacement on the cached content queue, the CDN edge server eliminates less popular cached content in the cached content queue according to the popularity information of the cached content in the cached content queue. 24. The method according to claim 23, wherein if the identification information of the content to be accessed is not in the cached content queue, the CDN edge server determines the content to be accessed according to the identification information of the content to be accessed Whether the accessed content is in the candidate content queue of the CDN edge server; If the content to be accessed is in the candidate content queue, the CDN edge server updates the access popularity of the content to be accessed, and determines the popularity of the content to be accessed according to the access popularity and recommendation popularity of the content to be accessed information; The CDN edge server determines whether the popularity of the content to be accessed is greater than a preset popularity threshold according to the popularity information of the content to be accessed; If the popularity of the content to be accessed is greater than the popularity threshold, the CDN edge server adds the content to be accessed to the cache content queue, and deletes the content to be accessed from the candidate content queue; The CDN edge server returns to the client the network protocol IP address of the original server where the content server to be accessed is located. 25. The method according to claim 24, wherein if the content to be accessed is not in the content candidate queue, the CDN edge server adds the content to be accessed to the content candidate queue; The CDN edge server updates the access popularity of the content to be accessed, determines the popularity information of the content to be accessed according to the popularity of the content to be accessed and the recommendation popularity of the content to be accessed, and determines the popularity information of the content to be accessed according to the popularity of the content to be accessed information to update the candidate content queue; The CDN edge server returns to the client the IP address of the original server where the content server to be accessed is located. 26. The method according to claim 24, wherein if the popularity of the content to be accessed is less than or equal to the popularity threshold, the CDN edge server updates the candidate according to the popularity information of the content to be accessed content queue; The CDN edge server returns to the client the IP address of the original server where the content server to be accessed is located. 27. The method according to any one of claims 24-26, further comprising: When it is necessary to cache and replace the candidate content queue, the CDN edge server eliminates less popular candidate content in the candidate content queue according to the popularity information of the candidate content in the candidate content queue. 28. A recommendation system, characterized in that it comprises: The receiving module is configured to receive the identifier of the cached content in the cached content queue sent by the content distribution network CDN, obtain the information of the cached content according to the identifier of the cached content and the entire content library, and combine the identifier of the cached content with the Add the information about the cached content to the cached content repository; A recommendation module, configured to use a first recommendation algorithm to calculate a first recommendation result according to the pre-acquired user interest characteristics and the entire content library when the recommendation system receives a recommendation request message sent by the client; The recommendation module is further configured to obtain a second recommendation result according to the cached content library; a fusion module, configured to fuse the first recommendation result and the second recommendation result according to a preset fusion algorithm to obtain a target recommendation result; A sending module, configured to push the target recommendation result to the target user. 29. The recommendation system according to claim 28, wherein the recommendation module obtains a second recommendation result according to the cached content library, specifically: According to the user interest feature and the cached content library, the second recommendation result is calculated by using a second recommendation algorithm. 30. The recommendation system according to claim 28, wherein the recommendation module acquires a second recommendation result according to the cached content library, specifically: Select recommended content belonging to the cached content library from the first recommended result, and use the selected recommended content as the second recommended result. 31. The recommendation system according to any one of claims 28-30, wherein the fusion module is specifically used for: determining common recommended content in the first recommendation result and the second recommendation result; deleting the common recommendation content from the first recommendation result to obtain a third recommendation result; According to the score of the recommended content, the recommended content in the second recommended result and the third recommended result is uniformly sorted; The sorted recommended content is used as the target recommendation result, or part of the recommended content is selected from the sorted recommended content according to a preset algorithm as the target recommendation result. 32. The recommendation system according to any one of claims 28-30, wherein the fusion module is specifically used for: determining common recommended content in the first recommendation result and the second recommendation result; deleting the common recommendation content from the first recommendation result to obtain a third recommendation result; Select a%*k recommended content from the third recommendation result, where k is the number of recommended content included in the target recommendation result, and a is greater than or equal to 0 and less than or equal to 100; Select (1-a%)*k recommended content from the second recommended result; According to the scores of the recommended content, the recommended content selected from the third recommended result and the recommended content selected from the second recommended result are uniformly sorted, and the uniformly sorted recommended content is used as the target recommended result. 33. The recommendation system according to claim 32, wherein the fusion module selects a%*k recommended content from the third recommendation result, specifically: sorting the recommended content in the third recommendation result according to the score of the recommended content, and selecting the top a%*k recommended content from the sorted third recommendation result; The fusion module selects (1-a%)*k recommended content from the second recommendation result, specifically: The recommended content in the second recommended result is sorted according to the score of the recommended content, and the top (1-a%)*k recommended content is selected from the sorted second recommended result. 34. The recommendation system according to any one of claims 31-33, wherein after the fusion module deletes the common recommendation content from the first recommendation result, it is further used for: The score of the common recommended content included in the second recommendation result is increased. 35. The recommendation system according to any one of claims 28-30, wherein the fusion module is specifically used for: determining common recommended content in the first recommendation result and the second recommendation result; deleting the common recommendation content from the second recommendation result to obtain a fourth recommendation result; According to the score of the recommended content, the recommended content in the first recommended result and the fourth recommended result is uniformly sorted; The sorted recommended content is used as the target recommendation result, or part of the recommended content is selected from the sorted recommended content according to a preset algorithm as the target recommendation result. 36. The recommendation system according to any one of claims 28-30, wherein the fusion module is specifically used for: determining common recommended content in the first recommendation result and the second recommendation result; deleting the common recommendation content from the second recommendation result to obtain a fourth recommendation result; Select a%*k recommended content from the first recommendation result, where k is the number of recommended content included in the target recommendation result, and a is greater than or equal to 0 and less than or equal to 100; Select (1-a%)*k recommended content from the fourth recommended result; According to the score of the recommended content, the recommended content selected from the first recommended result and the recommended content selected from the fourth recommended result are uniformly sorted, and the unified sorted recommended content is used as the target recommended result. . 37. The recommendation system according to claim 36, wherein the fusion module selects a%*k recommended content from the first recommendation result, specifically: sorting the recommended content in the first recommendation result according to the score of the recommended content, and selecting the top a%*k recommended content from the sorted first recommendation result; The fusion module selects (1-a%)*k recommended content from the fourth recommendation result, specifically: The recommended content in the fourth recommended result is sorted according to the score of the recommended content, and (1-a%)*k recommended content is selected from the sorted fourth recommended result. 38. The recommendation system according to any one of claims 35-37, wherein after the fusion module deletes the common recommendation content from the second recommendation result, it is further used for: The score of the common recommended content included in the first recommendation result is increased. 39. The recommendation system according to any one of claims 28-30, wherein the recommendation system further comprises: A recommendation popularity generation module, configured to generate a recommendation popularity library according to the recommendation situation of all contents in the entire content library, the recommendation popularity library including the recommendation popularity of all the contents in the entire content library within a preset time; The sending module is further configured to send all the content in the recommended popularity library to the CDN. 40. The recommendation system according to claim 39, wherein the fusion module fuses the first recommendation result and the second recommendation result according to a preset fusion algorithm, and after obtaining the target recommendation result, the The recommendation heat generating module is also used for: The recommendation popularity database is updated according to the target recommendation result. 41. The recommendation system according to any one of claims 28-41, wherein the cached content sent by the CDN is the content of the top P% of the cached content queue, or is the content of the cached content queue Incremental data of the content of the top P% relative to the content sent last time, where P is greater than 0 and less than 100. 42. A content distribution network CDN edge server, characterized in that it comprises: An acquisition module, configured to acquire recommendation popularity and access popularity of cached content in the cached content queue; A cache replacement module, configured to perform cache replacement on the cache content queue according to the access popularity and recommendation popularity of the cache content in the cache content queue. 43. The CDN edge server according to claim 42, wherein the cache replacement module is specifically used for: If the size of the cached content in the cached content queue is greater than or equal to the first threshold, it is determined to eliminate the cached content at the end of the cached content queue with less popular access; Comparing the recommendation heat of the cache content with the same access heat at the queue tail of the cache content queue, eliminating the cache content with the lower recommendation heat among the cache content with the same access heat until the cache content in the cache content queue reaches If the size is smaller than a second threshold, stop eliminating the cached content queue, and the second threshold is smaller than or equal to the first threshold. 44. The CDN edge server according to claim 42, wherein the cache replacement module is specifically used for: If the size of the cached content in the cached content queue is greater than or equal to the first threshold, it is determined to eliminate the cached content at the queue tail of the cached content queue with less popular access; According to the access heat and recommendation heat of the cache content at the queue tail of the cache content queue, calculate the comprehensive heat of the cache content in the queue tail of the cache content queue; Eliminate the cache content with less comprehensive heat in the queue tail of the cache content queue until the size of the cache content in the cache content queue is less than a second threshold, then stop eliminating the cache content queue, the second The threshold is less than or equal to the first threshold. 45. The CDN edge server according to claim 42, wherein the acquiring module is specifically used for: generating the access popularity of the cached content in the cached content queue according to the historical access conditions of the cached content in the cached content queue; receiving the recommendation heat of the cache content in the cache content queue sent by the recommendation system, the recommendation heat of the cache content in the cache content queue is generated by the recommendation system according to the recommendation situation of the cache content in the cache content queue . 46. The CDN edge server according to any one of claims 42-45, wherein the acquisition module is also used for: Obtain the recommendation popularity and access popularity of the candidate content in the candidate content queue; The cache replacement module is further configured to: perform cache replacement on the candidate content queue according to the recommendation popularity and access popularity of the candidate content in the candidate content queue. 47. The CDN edge server according to claim 46, wherein the cache replacement module is specifically used for: If the size of the candidate content in the candidate content queue is greater than or equal to a third threshold, it is determined to eliminate the candidate content at the tail of the candidate content queue that has less popular access; Comparing the recommendation popularity of the candidate content with the same access popularity in the candidate content queue, eliminating the candidate content with the lower recommendation popularity among the candidate contents with the same access popularity, until the size of the candidate content in the candidate content queue is less than the fourth threshold, stop eliminating the candidate content queue, and the fourth threshold is less than or equal to the third threshold. 48. The CDN edge server according to claim 46, wherein the cache replacement module is specifically used for: If the size of the candidate content in the candidate content queue is greater than or equal to a third threshold, it is determined to eliminate the candidate content at the tail of the candidate content queue that has less popular access; calculating the comprehensive popularity of the candidate content in the tail of the candidate content queue according to the access popularity and recommendation popularity of the candidate content in the queue tail of the candidate content queue; Eliminate the candidate content with less comprehensive popularity of the candidate content in the queue tail of the candidate content queue, until the size of the candidate content in the candidate content queue is less than the fourth threshold, then stop eliminating the candidate content queue, so The fourth threshold is less than or equal to the third threshold. 49. The CDN edge server according to claim 46, wherein the acquiring module is specifically used for: generating the access popularity of the candidate content in the candidate content queue according to the historical access conditions of the candidate content in the candidate content queue; The recommendation popularity of the candidate content sent by the recommendation system is received, and the recommendation popularity of the candidate content is generated by the recommendation system according to the recommendation situation of the candidate content. 50. A content distribution network CDN edge server, characterized in that it comprises: A receiving module, configured to receive a content acquisition request sent by the client, where the content acquisition request includes identification information of the content to be accessed; A processing module, configured to determine whether the content to be accessed is in its own cache content queue according to the identification information of the content to be accessed, and return to the client if the content to be accessed is in the cache content queue The content to be accessed; An update module, configured to update the access popularity of the content to be accessed, and calculate the popularity information of the content to be accessed according to the access popularity and recommendation popularity of the content to be accessed, and update the Cache content queue; The cache replacement module is configured to, when it is necessary to perform cache replacement on the cache content queue, eliminate less popular cache content in the cache content queue according to the popularity information of the cache content in the cache content queue. 51. The CDN edge server according to claim 50, wherein if the identification information of the content to be accessed is not in the cached content queue, the processing module is further configured to: determining whether the content to be accessed is in the candidate content queue of the CDN edge server according to the identification information of the content to be accessed; If the content to be accessed is in the candidate content queue, update the access popularity of the content to be accessed, and determine the popularity information of the content to be accessed according to the access popularity and recommendation popularity of the content to be accessed; judging whether the popularity of the content to be accessed is greater than a preset popularity threshold according to the popularity information of the content to be accessed; If the popularity of the content to be accessed is greater than the popularity threshold, adding the content to be accessed to the cache content queue, and deleting the content to be accessed from the candidate content queue; Returning the network protocol IP address of the original server where the content server to be accessed is located to the client. 52. The CDN edge server according to claim 51, wherein if the content to be accessed is not in the candidate content queue, the processing module is further configured to: adding the content to be accessed to the candidate content queue; Update the access popularity of the content to be accessed, determine the popularity information of the content to be accessed according to the popularity of the content to be accessed and the recommendation popularity of the content to be accessed, and update the candidate according to the popularity information of the content to be accessed content queue; Returning to the client the IP address of the original server where the content server to be accessed is located. 53. The CDN edge server according to claim 51, wherein if the popularity of the content to be accessed is less than or equal to the popularity threshold, the processing module is further configured to: updating the candidate content queue according to the popularity information of the content to be accessed; Returning to the client the IP address of the original server where the content server to be accessed is located. 54. The CDN edge server according to any one of claims 51-53, wherein the cache replacement module is also used for: When it is necessary to cache and replace the candidate content queue, the less popular candidate content in the candidate content queue is eliminated according to the popularity information of the candidate content in the candidate content queue.