CN104298719B - Category division, advertisement placement method and the system of user is carried out based on Social behaviors - Google Patents

Abstract

The invention discloses a kind of category division, advertisement placement method and system that user is carried out based on Social behaviors, this method includes：According to the UID of user to be sorted, obtain to should be in the setting cycle of UID records social daily record, and as the user social contact data of the user；Cookie ID according to corresponding to the UID, obtain to should be in the setting cycle of cookie ID records access log, and as the user behavior data of the user；User behavior attributive character is extracted from the user behavior data, user social contact attributive character is extracted from the user social contact data；According to the user behavior attributive character and user social contact attributive character of extraction, the user is divided into respective classes.Using the present invention, the degree of accuracy and the advertisement putting validity of the category division of user can be improved.

Description

Method and system for classifying user categories and putting advertisements based on social behaviors

Technical Field

The invention relates to the field of Internet, in particular to a method and a system for classifying users and putting advertisements based on social behaviors.

Background

With the continuous development of internet technology, internet advertisements rapidly replace traditional media advertisements with the advantages of rapidness, convenience and strong flexibility. The internet advertisement is a high-tech advertisement operation mode which puts advertisements on the internet through a network advertisement platform, places or publishes advertisements on the internet by using advertisement banners, text links and multimedia methods on websites, and transmits the advertisements to internet users through the network. Compared with the traditional four-media (newspaper, magazine, television and broadcast) advertisement and the outdoor advertisement which is prepared by the blue-green phenomenon, the internet advertisement takes the internet as the advertisement media, has the advantage of unique property, and is an important part for implementing the strategy of modern marketing media.

Moreover, with the rapid development of internet advertising technology, the demand of advertisers for target audiences is also increasing. The traditional media advertisement has short life cycle, complex manufacture, poor flexibility and timeliness and is easy to be interfered by external factors; more particularly, they cannot accurately lock the audience and people who need to be targeted, so that it is urgently needed to provide an advertisement mode which can be oriented to people and used for improving the popularization effect and greatly improving the popularization effect.

In practical applications, IGRP (Internet Rating Points) advertisements are advertisement selling schemes facing to people evolved from GRP (Rating Points) of traditional television advertisements. Characteristics such as age, gender, interest and the like of audience users can be deeply mined through the advantages of internet big data, and then the users are classified according to the mined characteristics, so that advertisers can put advertisements in a targeted manner, and the popularization effect is improved.

A user category classification method is provided in the prior art, which mainly classifies users into corresponding categories according to the access logs of the users at the advertisement delivery websites. Specifically, after an access log of a user on an advertisement delivery website is acquired, a page accessed by the user is analyzed to obtain characteristics of the page accessed by the user; then, the category of the user is inferred from the characteristics of the page visited by the user. However, in fact, the category of the user is deduced reversely according to the characteristics of the page visited by the unilateral user, which is not highly reliable, and results in low accuracy of the category classification result of the user, and then results in insufficient audience pertinence and poor popularization effect of the advertisement delivered according to the category classification result, and reduces effectiveness of advertisement delivery.

Therefore, it is necessary to provide a user classification method that improves the classification accuracy.

Disclosure of Invention

Aiming at the defects in the prior art, the invention provides a method and a system for classifying the categories of users and delivering advertisements based on social behaviors, which are used for improving the accuracy of classification of the users and the effectiveness of advertisement delivery.

According to one aspect of the invention, a method for classifying users based on social behaviors is provided, which comprises the following steps:

acquiring user behavior data of users to be classified in a set period and user social data;

extracting user behavior attribute features from the user behavior data, and extracting user social attribute features from the user social data;

and classifying the users into corresponding categories according to the extracted user behavior attribute characteristics and the extracted user social attribute characteristics.

Preferably, the classifying the users into corresponding categories according to the extracted user behavior attribute features and user social attribute features specifically includes:

matching the user behavior attribute characteristics of the user with the user behavior attribute characteristics recorded in a pre-established behavior characteristic rule base, and judging a behavior group to which the user belongs according to a matching result; the behavior feature rule base records user behavior attribute features of each behavior family for each behavior family;

matching the user social attribute features of the user with the user social attribute features recorded in a pre-established social feature rule base, and judging the social group to which the user belongs according to the matching result; the social feature rule base records user social attribute features of each social group;

and finding out a common group between the behavior group and the social group to which the user belongs, and dividing the user into the found common group.

Preferably, the classifying the users into corresponding categories according to the extracted user behavior attribute features and user social attribute features specifically includes:

fusing and summarizing the extracted user behavior attribute features and the extracted user social attribute features of the user to obtain user features of the user;

inputting the obtained user characteristics of the user into a category classification model for category classification; and classifying the users into corresponding categories through the category classification model.

Preferably, the class classification model is pre-trained:

for each training user selected from users of the website, after counting user behavior attribute features and user social attribute features of the training user, fusing and summarizing to obtain user features of the training user; determining a behavior group to which the training user belongs according to the user behavior attribute characteristics of the training user, determining a social group to which the training user belongs according to the user social attribute characteristics of the training user, and taking a common group between the behavior group and the social group to which the training user belongs as the group to which the training user belongs;

and taking the family group and the user characteristics of each training user as training data, and performing model training by using the training data and applying a preset multi-label multi-classification algorithm to obtain a class classification model.

Preferably, before the obtaining the category classification model, the method further includes:

for each test user selected from users of the website, after the user behavior attribute characteristics and the user social attribute characteristics of the test user are counted, the user characteristics of the test user are obtained through fusion and summarization; determining a behavior group to which the test user belongs according to the user behavior attribute characteristics of the test user, determining a social group to which the test user belongs according to the user social attribute characteristics of the test user, and taking a common group between the behavior group and the social group to which the test user belongs as the group to which the test user belongs;

taking the belonged group and the user characteristics of each test user as test data; and

and when the training data is used for carrying out model training by using a preset multi-label multi-classification algorithm, the test data is also used for carrying out model training by using a preset multi-label multi-classification algorithm to obtain the class classification model.

The invention also provides an advertisement putting method, which comprises the following steps:

after receiving an access request of a user, a website acquires an identity identification number (CookieID) of the user and a corresponding user identity certificate (UID);

obtaining a classification result of the user by adopting the method according to any one of claims 1 to 5 according to the cookie ID and UID of the user;

taking the obtained category division result as a search keyword, and finding out advertisement content matched with the search keyword;

and playing the searched advertisement content in an advertisement playing column of the webpage returned to the user aiming at the access request.

According to another aspect of the present invention, there is also provided a system for classifying a user based on social behaviors, including:

the user data acquisition module is used for acquiring and outputting user behavior data and user social data of users to be classified in a set period;

the attribute feature extraction module is used for extracting user behavior attribute features from the user behavior data output by the user data acquisition module and extracting user social attribute features from the user social data output by the user data acquisition module;

and the user category dividing module is used for dividing the users into corresponding categories according to the user behavior attribute characteristics and the user social attribute characteristics extracted by the attribute characteristic extraction module.

Preferably, the user category classification module specifically includes: a first user category dividing unit or a second user category dividing unit; wherein,

the first user category dividing unit specifically includes:

a behavior feature matching subunit, configured to match the user behavior attribute features extracted by the attribute feature extraction module with the user behavior attribute features recorded in a pre-established behavior feature rule base, and determine a behavior group to which the user belongs according to a matching result; the behavior feature rule base records user behavior attribute features of each behavior family for each behavior family;

the social characteristic matching subunit is used for matching the user social attribute characteristics extracted by the attribute characteristic extraction module with the user social attribute characteristics recorded in a pre-established social characteristic rule base and judging the social group to which the user belongs according to the matching result; the social feature rule base records user social attribute features of each social group;

the category dividing unit is used for receiving the behavior group to which the user belongs and output by the behavior feature matching unit and the social group to which the user belongs and output by the social feature matching unit; finding a common group between a behavior group and a social group to which the user belongs, and dividing the user into the found common group; and

the second user category dividing unit is used for fusing and summarizing the user behavior attribute features and the user social attribute features of the users extracted by the attribute feature extraction module to obtain the user features of the users; inputting the obtained user characteristics of the user into a category division model for category division; and classifying the users into corresponding categories through the category classification model.

Preferably, the system for classifying the user based on the social behavior further comprises:

the system comprises a division model training module, a classification model selection module and a classification model selection module, wherein the division model training module is used for carrying out fusion and summarization to obtain the user characteristics of a training user after the user behavior attribute characteristics and the user social attribute characteristics of the training user are counted for each training user selected from users of a website; determining a behavior group to which the training user belongs according to the user behavior attribute characteristics of the training user, determining a social group to which the training user belongs according to the user social attribute characteristics of the training user, and taking a common group between the behavior group and the social group to which the training user belongs as the group to which the training user belongs; and taking the family group and the user characteristics of each training user as training data, and performing model training by using the training data and applying a preset multi-label multi-classification algorithm to obtain a class classification model.

The invention also provides an advertisement delivery system, comprising:

the system comprises a user data acquisition module, an attribute feature extraction module and a user category division module in the system for classifying users based on social behaviors;

the webpage request receiving module is used for acquiring the cookie ID and the corresponding UID of the user after receiving the access request of the user and sending the cookie ID and the corresponding UID to the user data acquiring module;

the advertisement query module is used for taking the classification result of the user output by the user classification module as a search keyword and searching the advertisement content matched with the search keyword;

the request processing module is used for returning the webpage to the user after inserting the advertisement content found by the advertisement inquiry module into the advertisement playing column of the webpage requested by the access request;

the user data acquisition module is specifically used for acquiring and outputting user behavior data and user social data of the user in a set period according to the cookie ID and the corresponding UID of the user.

According to the technical scheme, user behavior attribute features and user social attribute features are respectively extracted according to user behavior data of a user on a website and user social data under a social platform; and then, classifying the user according to the extracted user behavior attribute characteristics and the extracted user social attribute characteristics. Compared with the prior art that the interest of the user is reversely deduced only according to the information of the user access page, the technical scheme provided by the invention comprehensively considers the user behavior attribute characteristics and the user social attribute characteristics to classify the user category, thereby increasing the feature richness of the category classification, improving the accuracy of the user category classification and improving the positioning accuracy of the crowd; therefore, based on higher crowd positioning accuracy, the pertinence and the delivery effectiveness of audiences who advertise and publicize videos and the like can be improved.

Drawings

FIG. 1 is a flowchart illustrating a method for classifying users based on social behaviors according to an embodiment of the present invention;

FIG. 2 is a flowchart illustrating an advertisement delivery method according to an embodiment of the present invention;

FIG. 3 is a schematic flow chart illustrating the process of extracting user behavior attribute features according to an embodiment of the present invention;

FIG. 4 is a flowchart illustrating a process of extracting social attribute features of a user according to an embodiment of the present invention;

FIG. 5 is a schematic flowchart of a training method of a class classification model according to an embodiment of the present invention;

FIG. 6 is a schematic structural diagram of a system for classifying users based on social behaviors according to an embodiment of the present invention;

fig. 7 is a schematic structural diagram of a first user category dividing unit according to an embodiment of the present invention;

fig. 8 is a schematic structural diagram of an advertisement delivery system according to an embodiment of the present invention.

Detailed Description

The technical solutions of the present invention will be described clearly and completely with reference to the accompanying drawings, and it is obvious that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

As used in this application, the terms "module," "system," and the like are intended to include a computer-related entity, such as but not limited to hardware, firmware, a combination of hardware and software, or software in execution. For example, a module may be, but is not limited to: a process running on a processor, an object, an executable, a thread of execution, a program, and/or a computer. For example, an application running on a computing device and the computing device may both be a module. One or more modules may reside within a process and/or thread of execution and a module may be localized on one computer and/or distributed between two or more computers.

The inventor of the invention finds that, in a website, in addition to the condition that a user visits a page on the website, the interest of the user can be reflected to a certain degree, and the interest of the user can also be reflected to a certain degree by the blog articles issued, forwarded and complied on the social platform logged in by the user.

Therefore, the inventor of the present invention considers that the access log of the user on the website can be used as the user behavior data, the social log of the user under the social platform can be used as the user social data, and the user behavior attribute characteristics and the user social attribute characteristics are respectively extracted from the user behavior data and the user social data; then, the classification of the user can be carried out according to the extracted user behavior attribute characteristics and the extracted user social attribute characteristics. Compared with the prior art that the interest of the user is reversely deduced only according to the information of the user access page, the technical scheme provided by the invention comprehensively considers the user behavior attribute characteristics and the user social attribute characteristics to classify the user category, so that the feature richness of the category classification can be increased, the user category classification accuracy is improved, and the crowd positioning accuracy can also be improved; based on higher crowd positioning accuracy, the pertinence and the delivery effectiveness of audiences who advertise and publicize videos and the like can be improved.

The technical scheme of the invention is explained in detail in the following with the accompanying drawings.

The embodiment of the invention provides a method for classifying users based on social behaviors, wherein the flow is shown in fig. 1, and the method specifically comprises the following steps:

s101: and acquiring user behavior data of the user to be classified in a set period and user social data.

In practical applications, when a user sends an access request to a website, the website usually marks the user with a cookie ID (Identity) and records all access logs of the user. Correspondingly, if the User registers on the social platform of the website, a unique User Identification (UID) is obtained; in this way, each time the user logs into the social platform through the UID, the website may record the cookie id of the user and all social logs corresponding to the UID. The social platform in the embodiment of the invention can be microblog, twitter and the like.

Therefore, specifically, after the UID of the user to be classified is determined, the social log in the set period corresponding to the UID record may be obtained according to the UID of the user, and the social log may be used as the user social data of the user. Further, according to the cookie ID corresponding to the UID, an access log in a set period corresponding to the cookie ID record may be acquired and used as the user behavior data of the user.

The access log of the user in the set period may specifically include: each page visited by the user on the website in a set period and the number of visits, each advertisement clicked and the number of clicks, etc. The social log of the user in the set period may specifically include: user blog, user registration information, user interest information, Location Based Service (LBS) Location information of the user, and the like.

S102: and extracting the user behavior attribute characteristics from the user behavior data, and extracting the user social attribute characteristics from the user social data.

Specifically, according to the obtained user behavior data, the user behavior attribute features of the user, such as the gender, age, interest, etc., of the user, may be extracted therefrom. Further, according to the obtained user social data, user social attribute features of the user, such as gender, age, blog feature, attention user feature, LBS feature, and the like, can be extracted. In practical application, the user behavior attribute features extracted from the user behavior data may also be referred to as features based on cookie ID; the user social attribute features extracted from the user social data may also be referred to as UID-based features.

The following will describe how to extract the user behavior attribute features from the user behavior data and how to extract the user social attribute features from the user social data in detail.

S103: and classifying the users into corresponding categories according to the extracted user behavior attribute characteristics and the extracted user social attribute characteristics.

Specifically, the user behavior attribute features of the user may be matched with the user behavior attribute features recorded in the pre-established behavior feature rule base, and the behavior group to which the user belongs may be determined according to the matching result. The behavior feature rule base records the user behavior attribute features of each behavior group. Correspondingly, the obtained user social attribute features of the user can be matched with the user social attribute features recorded in the pre-established social feature rule base, and the social group to which the user belongs can be judged according to the matching result. The social characteristic rule base records the user social attribute characteristics of each social group.

In the embodiment of the present invention, the behavior group in the behavior feature rule base and the social group in the social feature rule base are preset, and may include: investment financing group, sports fitness group, urban white collar group, family group, youth campus group, movie and television group, photography group, etc.

Consider that a user will typically show similarities in access browsing behavior and social behavior if he is interested in something. Therefore, a common group between the behavior group to which the user belongs and the social group can be searched, and if the common group between the behavior group to which the user belongs and the social group is searched, the user can be classified into the searched common group. Compared with the existing categories divided based on unilateral behaviors, the category division method and the category division device have the advantages that the categories which are respectively judged through the characteristics based on the cookie ID and the UID and are common between the behavior groups and the social groups are used as the category division result of the user, and the accuracy is higher.

In practical application, the extracted user behavior attribute features can be matched with the user behavior attribute features of the behavior family group aiming at each behavior family group in the behavior feature rule base, and if the matching is successful, the behavior family group can be judged to be the behavior family group to which the user belongs; if not, it can be determined that the user does not belong to the behavior group. Correspondingly, the extracted user social attribute feature can be matched with the user social attribute feature of the social group aiming at each social group in the social feature rule base, and if the matching is successful, the social group can be judged to be the social group to which the user belongs; if not, it may be determined that the user does not belong to the social group.

Further, for each behavior group in the behavior feature rule base, the user behavior attribute features of the behavior group may specifically include user behavior attribute features that a plurality of users belonging to the behavior group must have, or may include user behavior attribute features that some users belonging to the behavior group commonly have.

Therefore, when the extracted user behavior attribute features are matched with the user behavior attribute features of the behavior group, if the extracted user behavior attribute features include the user behavior attribute features which the users belonging to the behavior group must have, the matching success can be judged; otherwise, the matching is judged to be unsuccessful. Accordingly, for each social group in the social feature rule base, the user social attribute features of the social group may specifically include user social attribute features that must be possessed by a plurality of users belonging to the social group, and may also include user social attribute features that are commonly possessed by some users belonging to the social group. Therefore, when the extracted user social attribute features are matched with the user social attribute features of the social group, if the extracted user social attribute features include the user social attribute features which the users belonging to the social group must have, the matching can be judged to be successful; otherwise, the matching is judged to be unsuccessful.

In addition, if there is no common clan between the behavior clan to which the user belongs and the social clan, the extracted user behavior attribute features and the extracted user social attribute features of the user can be fused and summarized to obtain the user features of the user. And then, classifying the users into corresponding categories according to the obtained user characteristics of the users and the pre-trained category classification model. Specifically, a plurality of features of similar meaning may be fused into one user feature; taking the feature without the similar meaning as a user feature; and fusing the contradictory characteristics into a user characteristic according to a preset fusion rule. For example, if the age feature in the extracted user behavior attribute features is inconsistent with the age feature in the extracted user social attribute features, the age feature in the user social attribute features may be directly used as the age feature in the user features according to a preset fusion rule about the age.

The present invention also provides a more preferred embodiment, and the step S102 is performed: after the user behavior attribute features are extracted from the user behavior data and the user social attribute features are extracted from the user social data, the extracted user behavior attribute features and the extracted user social attribute features of the user can be directly fused and summarized to obtain the user features of the user; then, the obtained user characteristics of the user can be input into a pre-trained class division model for class division; and classifying the users into corresponding categories through a category classification model.

In the embodiment of the present invention, the training of the class classification model will be described in detail later.

In practical application, after the users are classified into corresponding categories according to the extracted user behavior attribute features and the extracted user social attribute features, the categories classified by the users can be stored as category classification results. In consideration of the fact that the user logs in websites at different computers in practical application, the cookie IDs recorded by the websites corresponding to the user are different, that is, a plurality of cookie IDs may exist corresponding to the same user; but a unique UID exists for the same user; therefore, further, the classification result of the user can be stored in correspondence with the UID of the user, which is convenient for subsequent use. Of course, in actual application, the cookie ID of the user may be stored in association with the UID and the classification result.

Based on the category classification method, the invention further provides an advertisement delivery method, the flow of which is shown in fig. 2, and the method specifically includes the following steps:

s201: after receiving an access request of a user, the website acquires a cookie ID and a corresponding UID of the user.

Specifically, after receiving an access request of a user, the website can normally acquire a webpage corresponding to the access request; the cookie ID of the user and its corresponding UID may also be obtained.

S202: and acquiring user behavior data of the user in a set period and user social data according to the acquired cookie ID and UID.

S203: extracting user behavior attribute features from the obtained user behavior data, and extracting user social attribute features from the obtained user social data; and classifying the users into corresponding categories according to the extracted user behavior attribute characteristics and the extracted user social attribute characteristics to obtain category classification results of the users.

Preferably, after the website receives the access request of the user and acquires the cookie ID and the UID of the user, the website can directly find out the category classification result corresponding to the acquired cookie ID and UID from the user category classification result library as the category classification result of the user. Wherein, the user classification result base stores the classification result divided according to the method as steps S101-S103 and the cookie ID and UID corresponding to the classification result for each user.

S204: using the obtained classification result as a search keyword, and finding out advertisement content matched with the search keyword; and playing the searched advertisement content in an advertisement playing column of the webpage returned to the user aiming at the access request.

Specifically, after the search keyword is obtained, the advertisement content matched with the search keyword can be found from the background of the website by a method well known to those skilled in the art; and adding all the acquired advertisement contents to an advertisement playing bar of a webpage returned to the user aiming at the access request. Therefore, when the user visits the webpage of the website, the website can push advertisements which are possibly interested by the user to the user according to the classification result of the user, and the advertisement putting accuracy is improved.

In the embodiment of the invention, before classifying the users, each page attribute feature and the score of each page attribute can be correspondingly stored for each page of the website in advance according to the content of the page and the characteristics of audience users.

The page attributes may specifically include: age, gender and hobbies; each page attribute has different page attribute characteristics. Thus, for each page, the page may have a corresponding score on different page attribute characteristics, depending on the characteristics of the audience user of the page. For convenience of description, in the embodiment of the present invention, the score of the page attribute feature may also be referred to as a page attribute feature score.

In practical applications, the page attribute feature of the page attribute-age (also referred to as age feature for short) may specifically include: 10-20 years old, 21-30 years old, 31-40 years old, or 41 years old or more. For example, for a certain page of a website, the score of the page attribute feature-10 to 20 years old corresponding to the page is 0.1; the page attribute characteristics corresponding to the page are 0.7 in the score of-21 to 30 years old; the score of the page attribute characteristics of-31 to 40 years old corresponding to the page is 0.1; the page attribute feature corresponding to the page-a score of 0.1 for 41 years old or older.

Page attribute-gender page attribute features (also referred to simply as gender features) may specifically include: male and female. For example, for a certain page of the website, the score of male, which is the page attribute feature corresponding to the page, is 0.3; the page attribute feature corresponding to this page-female score is 0.7.

Page attribute-page attribute features of interest (also referred to simply as interest features) may specifically include; reading, traveling, real estate, second-hand house, basketball, football, financing, etc. For example, for a certain page of the website, the score of the page attribute feature-financing corresponding to the page is 0.5; the score of the page attribute characteristic-reading corresponding to the page is 0.3; the score of the property characteristic-property of the page corresponding to the page is 0.2; and the scores of the page attribute corresponding to the page, namely other page attribute characteristics under the interest are all 0.

Based on the page attribute features and their scores corresponding to the pre-stored pages, as shown in fig. 3, how to extract the user behavior attribute features from the user behavior data in step S102 may be specifically extracted through the following steps:

s301: and determining each page accessed by the user in a set period from the acquired user behavior data.

Specifically, in addition to determining each page that the user accesses in the set period, the user behavior data may also determine the number of accesses of each page.

S302: and aiming at each page visited by the user in a set period, acquiring each page feature and the score of the page under each page attribute.

S303: and determining the page characteristics of the user under each page attribute according to the scores of all the page attribute characteristics of each page under each page attribute, and taking the determined page characteristics as the extracted user behavior attribute characteristics.

Specifically, for a page attribute-age, for each page attribute feature (age feature) under the page attribute, an average value of scores of the page attribute features may be calculated according to the obtained score of the page attribute feature stored corresponding to each page; and selecting the page attribute feature with the highest average value as the page feature of the user under the page attribute, and as the extracted user behavior attribute feature. Correspondingly, for the page attribute-gender, for each page attribute feature (gender feature) under the page attribute, calculating the average value of the scores of the page attribute features according to the scores of the page attribute features stored corresponding to the acquired pages; and selecting the page attribute feature with the highest average value as the page feature of the user under the page attribute, and as the extracted user behavior attribute feature.

Further, after acquiring each page attribute feature and its score under the page attribute-interest corresponding to each page visited by the user, accumulating the scores of the page attribute features corresponding to each page and stored under the page attribute-interest, and if the accumulated score of the page attribute features is greater than a set page attribute score threshold, taking the page attribute features as the page features of the user under the page attribute-interest and as the extracted user behavior attribute features. Wherein, the page attribute score threshold may be specifically set to 0.

In practical application, relative to the interests of the user, the gender and age of the user belong to the basic attributes of the user, so in the embodiment of the present invention, the user behavior attribute characteristics of the user can be divided into: basic attribute feature based on behaviors, interest feature based on behaviors. Thus, the page features of the user under the page attribute-age and the page features under the page attribute-gender specifically belong to the behavior-based basic attribute features in the user behavior attribute features of the user; the page features of the user under the page attribute-interest specifically belong to the interest features based on the behaviors in the user behavior attribute features of the user.

In practical application, if a user is interested in a certain channel classified in advance by a website, the number of access times of pages of the user under the channel is large; therefore, preferably, in the embodiment of the present invention, while performing step S303, the user behavior attribute feature may also be extracted through the following steps:

s304: and counting all the access channels of the user on the website according to the channels to which each page accessed by the user in a set period belongs, and taking the counted access channels as the extracted user behavior attribute characteristics.

Specifically, determining a channel to which each page accessed by a user in a set period belongs; and counting all access channels on the website in a user set period according to the determined channels to which the pages belong, and taking the counted access channels as one of interest characteristics based on behaviors in the user behavior attribute characteristics of the user. The channels of each page on the website are divided in advance.

Further, for each counted access channel, the feature score of the access channel may be determined according to the number of accesses to each page belonging to the access channel in the period set by the user. In this way, more preferably, the access channels with the feature scores not exceeding the set access channel feature score threshold can be eliminated from the behavior attribute features of the user, and the accuracy of the classification result classified based on the behavior attribute features of the user is ensured.

For example, for each visited channel, the sum of the number of visits of each page belonging to the visited channel in the user-set period may be used as the feature score of the channel feature. Or, for each access channel, the sum of the access times of each page belonging to the access channel in the user setting period may be used as the access frequency of the access channel; and taking the ratio of the access frequency of the channel characteristic to the sum of the access frequencies of all the counted access channels as the characteristic score of the channel characteristic.

In practical application, in addition to determining each page and the access times thereof accessed by the user in the set period from the user behavior data, each advertisement clicked by the user in the set period and the click times thereof can also be determined.

Therefore, preferably, in the embodiment of the present invention, while performing step S303, the user behavior attribute feature may also be extracted through the following steps:

s305: and counting each clicking industry of the user on the website according to the industry to which each advertisement clicked on the website by the user in a set period belongs, and taking the counted clicking industry as the extracted user behavior attribute characteristics.

Specifically, aiming at each advertisement clicked on a website by a user in a set period, determining the industry to which the advertisement belongs; and counting all click industries on the website in a user set period according to the determined industries to which the advertisements belong, and taking the counted click industries as one of interest characteristics based on behaviors in the user behavior attribute characteristics of the user. The industry to which each advertisement on the website belongs is divided in advance.

Further, for each counted click industry, the feature score of the click industry can be determined according to the click times of the advertisements belonging to the click industry in the period set by the user. Therefore, the clicking industry with the characteristic score not exceeding the set clicking industry characteristic score threshold can be removed from the user behavior attribute characteristics of the user, and the accuracy of the classification result of the classification of the user behavior attribute characteristics is guaranteed.

For example, for each click industry, the sum of the click times of the advertisements belonging to the click industry in the user-set period may be used as the feature score of the click industry; or, for each click industry, the sum of the click times of the advertisements belonging to the click industry in the period set by the user may be used as the click frequency of the click industry, and the ratio of the click frequency of the click industry to the sum of the click frequencies of all the click industries may be used as the feature score of the click industry.

As to how to extract the user social attribute feature from the user social data mentioned in the above step S102, as shown in fig. 4, the user social attribute feature may be extracted specifically through the following steps:

s401: and determining each blog article of the user in a set period from the obtained social data of the user.

The blog article of the user may specifically include: and the user initiates, forwards, reviews and approves the blog article.

S402: extracting subject terms of the determined blog articles from each blog article; and taking all the extracted blog article subject terms of the blog articles as the social attribute features of the users.

In the embodiment of the present invention, the social attribute features of the user can be divided into: social-based basic attribute features, social-based interest features. The extracted blog article subject term specifically belongs to one of the user social attribute features of the user based on social interest features. The conventional technical means of those skilled in the art can be adopted to extract the subject words of the Bo-Wen, and the details are not described herein.

Further, for each extracted subject word of the blog, a feature score of the subject word of the blog can be determined according to the occurrence frequency of the subject word of the blog in all the blog. In practical application, for each blog article subject term, the counted occurrence frequency of the blog article subject term in all the blog articles can be directly used as the characteristic score of the blog article subject term. Or, the ratio of the occurrence number of the blog subject term in all the blog articles to the sum of the occurrence number of each blog subject term in all the blog articles may be used as the feature score of the blog subject term. Therefore, the blog subject term with the feature score not exceeding the set blog subject term feature score threshold can be removed from the user social attribute features of the user, and the accuracy of the classification result classified based on the user social attribute features is guaranteed.

In practical applications, a user needs to register before logging on to a social platform in a website, and the registration information of the user may generally include: age, gender, academic history, professional information, etc. of the user.

Therefore, more preferably, in the embodiment of the present invention, when performing steps S401 and S402, the user social attribute feature may also be extracted through the following steps:

s403: determining registration information of the user from the obtained social data of the user; and extracting the user social attribute characteristics of the user from the registration information of the user.

Specifically, the information of the age, sex, academic calendar, and the like of the user extracted from the registration information of the user can be directly used as the social-based basic attribute feature in the user social attribute features of the user.

Further, the user can select the custom tag on the social platform according to the situation of the user. It is considered that the user-defined tags can reflect the interests of the user to a certain extent. Therefore, further, in the embodiment of the present invention, while performing steps S401 and S402, the user social attribute feature may also be extracted through the following steps:

s404: and determining a user-defined label of the user from the obtained social data of the user, and taking the determined user-defined label as the extracted social attribute feature of the user.

Specifically, the user-defined tag determined from the obtained user social data may be used as a social-based interest feature in the user social attribute features of the user.

In practical applications, a user usually focuses on other users on a social platform, and objects focused on by the user can reflect the interests of the user to some extent. Therefore, more preferably, in the embodiment of the present invention, while performing steps S401 and S402, the user social attribute feature of the user may also be extracted according to the attention information of the user, and specifically, the user social attribute feature may be extracted through the following steps:

the social log of the user in the set period may specifically include: user blog, user registration information, user interest information, Location Based Service (LBS) Location information of the user, and the like.

S405: determining attention information of the user from the obtained social data of the user, and extracting an authenticated user which is concerned by the user and belongs to a specific industry from the determined attention information; and aiming at each extracted authentication user, taking the industry to which the authentication user belongs as the social attribute feature of the extracted user.

Specifically, the industry to which each authenticated user concerned by the user belongs may be taken as a social-based interest feature in the user social attribute features of the user.

For example, if the user pays attention to authenticated users belonging to the real estate industry, such as @ rexoplasma, @ new wave real estate, and @ beijing hometown, the real estate industry can be used as the extracted social attribute feature of the user.

Further, after determining the industry to which the authenticated user belongs for each extracted authenticated user, the industry concerned by the user can be counted according to the industry to which each authenticated user belongs; and aiming at the counted industries concerned by each user, calculating the characteristic score of the industry according to the number of the authenticated users belonging to the industry in the concerned information of the user. In this way, industries with characteristic scores not exceeding a set industry characteristic score threshold can be preferably eliminated from the user social attribute characteristics of the user, and the accuracy of classification results classified based on the user social attribute characteristics is guaranteed.

In practical applications, some geographic areas have obvious features (feature areas), such as national stadium, tourist attraction areas, business areas, etc.; if the user frequently goes in and out of the place, the interest of the user can be reflected to a certain degree. Therefore, it is possible to perform area division in advance, take these areas having distinct features as feature areas, and store the associated interests corresponding to the feature areas.

The inventor of the invention considers that users of the website often log in a social platform at a certain position based on certain factors (such as work, interests and the like); therefore, it may be considered that the user social attribute feature is extracted according to the LBS location of the user and the pre-divided feature area.

Specifically, because the LBS service of the social platform can record the login position of the user, the LBS position in the user setting period to be located can be directly obtained from the social log. And then, aiming at each obtained LBS position, according to the characteristic area to which the LBS position belongs, taking the interest associated with the characteristic area as the interest characteristic based on social contact in the user social attribute characteristic of the user.

In the embodiment of the present invention, the class classification model mentioned in step S103 is trained in advance, and as shown in fig. 5, the training may be specifically performed through the following steps:

s501: for each training user selected from users of the website, after the user behavior attribute features and the user social attribute features of the training user are counted, the user features of the training user are obtained through fusion and summarization.

S502: and for each training user, determining a behavior group to which the training user belongs according to the user behavior attribute characteristics of the training user.

S503: and for each training user, determining the social group to which the training user belongs according to the user social attribute characteristics of the training user.

S504: for each training user, a common clan between the behavior clan and the social clan to which the training user belongs is taken as the clan to which the training user belongs.

S505: and taking the family group and the user characteristics of each training user as training data, and performing model training by using the training data and applying a preset multi-label multi-classification algorithm to obtain a class classification model.

The preset multi-label multi-classification algorithm may be specifically an M3l algorithm or a Multiboost algorithm.

In the embodiment of the present invention, steps S501 to S504 may specifically refer to steps S101 to S103 in the method for classifying the user category based on the social behaviors, and are not described in detail here.

Preferably, in order to ensure that a class classification model with higher accuracy is obtained and improve the accuracy of a class classification result of a user, when training data is used for model training by using a preset multi-label multi-classification algorithm, test data can be further used for model training by using a preset multi-label multi-classification algorithm to obtain the class classification model. In practical application, the classification effect (such as accuracy and recall ratio) of the classification model can be evaluated by using the test data, so as to obtain the classification model with the optimal effect. The test data can be obtained specifically by:

for each test user selected from users of the website, after the user behavior attribute characteristics and the user social attribute characteristics of the test user are counted, the user characteristics of the test user are obtained through fusion and summarization; determining a behavior group to which the test user belongs according to the user behavior attribute characteristics of the test user, determining a social group to which the test user belongs according to the user social attribute characteristics of the test user, and taking a common group between the behavior group and the social group to which the test user belongs as the group to which the test user belongs; and taking the belonged group and the user characteristics of each test user as test data.

In practical application, in order to improve the model training efficiency, the training data of all training users may be preprocessed, which specifically includes: encoding, normalizing, dimension reduction and the like.

According to the method for classifying users based on social behaviors, an embodiment of the present invention further provides a system for classifying users based on social behaviors, which, as shown in fig. 6, may specifically include: a user data acquisition module 601, an attribute feature extraction module 602, and a user category classification module 603.

The user data obtaining module 601 is configured to obtain and output user behavior data and user social data of a user to be classified in a set period.

Specifically, the user data obtaining module 601 may obtain, according to the UID of the user to be classified, a social log in a set period corresponding to the UID record, and use the social log as the user social data of the user. Further, according to the cookie ID corresponding to the UID, an access log in a set period corresponding to the cookie ID record may be acquired and used as the user behavior data of the user.

The attribute feature extraction module 602 is configured to extract a user behavior attribute feature from the user behavior data output by the user data acquisition module 601, and extract a user social attribute feature from the user social data output by the user data acquisition module 601.

The user category classification module 603 is configured to classify the users into corresponding categories according to the user behavior attribute features and the user social attribute features extracted by the attribute feature extraction module 602.

Preferably, in the embodiment of the present invention, the system for classifying the categories of the user based on the social behavior further includes: the model training module 604 is partitioned.

The partition model training module 604 is configured to, for each training user selected from users of a website, perform statistics on user behavior attribute features and user social attribute features of the training user, and then perform fusion and summarization to obtain user features of the training user; determining a behavior group to which the training user belongs according to the user behavior attribute characteristics of the training user, determining a social group to which the training user belongs according to the user social attribute characteristics of the training user, and taking a common group between the behavior group and the social group to which the training user belongs as the group to which the training user belongs; and taking the family group and the user characteristics of each training user as training data, and performing model training by using the training data and applying a preset multi-label multi-classification algorithm to obtain a class classification model.

In this embodiment of the present invention, the user category dividing module 603 may specifically include: a first user category dividing unit or a second user category dividing unit.

The first user category classification unit is configured to classify users into corresponding categories according to the user behavior attribute features, the user social attribute features, the pre-established behavior feature rule base and the social feature rule base extracted by the attribute feature extraction module 602.

The second user category dividing unit is configured to fuse and summarize the user behavior attribute features and the user social attribute features of the user extracted by the attribute feature extraction module 602 to obtain user features of the user; inputting the obtained user characteristics of the user into a category division model for category division; and classifying the users into corresponding categories through a category classification model.

Further, as shown in fig. 7, the first user category dividing unit may specifically include: a behavior feature matching subunit 701, a social feature matching subunit 702, and a category classification subunit 703.

The behavior feature matching subunit 701 is configured to match the user behavior attribute features extracted by the attribute feature extraction module 602 with the user behavior attribute features recorded in the pre-established behavior feature rule base, and determine a behavior group to which the user belongs according to a matching result. The behavior feature rule base records the user behavior attribute features of each behavior group.

The social feature matching subunit 702 is configured to match the user social attribute features extracted by the attribute feature extraction module 602 with the user social attribute features recorded in the pre-established social feature rule base, and determine, according to a matching result, a social group to which the user belongs. The social characteristic rule base records the user social attribute characteristics of each social group.

The category matching subunit 703 is configured to receive the behavior group to which the user belongs, output by the behavior feature matching subunit 701, and the social group to which the user belongs, output by the social feature matching subunit 702; and finding out a common group between the behavior group and the social group to which the user belongs, and dividing the user into the found common group.

In the embodiment of the present invention, based on the advertisement delivery method and the user category division system, an embodiment of the present invention further provides an advertisement delivery system, as shown in fig. 8, which may specifically include: the system comprises a user data acquisition module 601, an attribute feature extraction module 602, a user category classification module 603, a webpage request receiving module 801, an advertisement query module 802 and a request processing module 803 in the system for classifying users based on social behaviors.

The web page request receiving module 801 is configured to, after receiving an access request of a user, obtain a cookie id and a corresponding UID of the user, and send the cookie id and the corresponding UID to the user data obtaining module 601. The user data obtaining module 601 may obtain and output user behavior data of the user and user social data in a set period according to the cookie id of the user and the corresponding UID.

The advertisement query module 802 is configured to use the category classification result of the user output by the user category classification module 603 as a search keyword, and find out advertisement content matching the search keyword.

The request processing module 803 is configured to, after inserting the advertisement content found by the advertisement querying module 802 into the advertisement playing bar of the webpage requested by the access request, return the webpage requested by the access request to the user.

In the embodiment of the present invention, the specific function implementation of each module, unit, and sub-unit in the system for classifying the user category based on the social behavior may refer to the detailed processes of steps S101 to 103, steps S301 to S305, and steps S401 to S405 in the method for classifying the user category based on the social behavior; the specific functional implementation of each module in the advertisement delivery system may refer to steps S201-204 in the advertisement delivery method described above, and will not be described in detail here.

According to the technical scheme, user behavior attribute features and user social attribute features are respectively extracted according to user behavior data of a user on a website and user social data under a social platform; and then, classifying the user according to the extracted user behavior attribute characteristics and the extracted user social attribute characteristics. Compared with the prior art that the interest of the user is reversely deduced only according to the information of the user access page, the technical scheme provided by the invention comprehensively considers the user behavior attribute characteristics and the user social attribute characteristics to classify the user category, so that the feature richness of the category classification can be increased, the user category classification accuracy is improved, and the crowd positioning accuracy can also be improved; based on higher crowd positioning accuracy, the pertinence and the delivery effectiveness of audiences who advertise and publicize videos and the like can be improved.

The foregoing is only a preferred embodiment of the present invention, and it should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be construed as the protection scope of the present invention.

Claims (5)

1. A method for classifying users based on social behaviors is characterized by comprising the following steps:

acquiring user behavior data of users to be classified in a set period and user social data;

extracting user behavior attribute features from the user behavior data, and extracting user social attribute features from the user social data;

classifying the users into corresponding categories according to the extracted user behavior attribute features and the extracted user social attribute features; the classifying the users into corresponding categories according to the extracted user behavior attribute features and the extracted user social attribute features specifically comprises:

matching the user behavior attribute characteristics of the user with the user behavior attribute characteristics recorded in a pre-established behavior characteristic rule base, and judging a behavior group to which the user belongs according to a matching result; the behavior feature rule base records user behavior attribute features of each behavior family for each behavior family;

matching the user social attribute features of the user with the user social attribute features recorded in a pre-established social feature rule base, and judging the social group to which the user belongs according to the matching result; the social feature rule base records user social attribute features of each social group;

finding a common group between a behavior group and a social group to which the user belongs, and dividing the user into the found common group; or the classifying the users into corresponding categories according to the extracted user behavior attribute features and the extracted user social attribute features specifically comprises:

fusing and summarizing the extracted user behavior attribute features and the extracted user social attribute features of the user to obtain user features of the user;

inputting the obtained user characteristics of the user into a category classification model for category classification; classifying the users into respective categories through the category classification model;

wherein the class classification model is pre-trained:

for each training user selected from users of the website, after counting user behavior attribute features and user social attribute features of the training user, fusing and summarizing to obtain user features of the training user; determining a behavior group to which the training user belongs according to the user behavior attribute characteristics of the training user, determining a social group to which the training user belongs according to the user social attribute characteristics of the training user, and taking a common group between the behavior group and the social group to which the training user belongs as the group to which the training user belongs;

and taking the family group and the user characteristics of each training user as training data, and performing model training by using the training data and applying a preset multi-label multi-classification algorithm to obtain a class classification model.

2. The method of claim 1, prior to said deriving a classification model, further comprising:

for each test user selected from users of the website, after the user behavior attribute characteristics and the user social attribute characteristics of the test user are counted, the user characteristics of the test user are obtained through fusion and summarization; determining a behavior group to which the test user belongs according to the user behavior attribute characteristics of the test user, determining a social group to which the test user belongs according to the user social attribute characteristics of the test user, and taking a common group between the behavior group and the social group to which the test user belongs as the group to which the test user belongs;

taking the belonged group and the user characteristics of each test user as test data; and

and when the training data is used for carrying out model training by using a preset multi-label multi-classification algorithm, the test data is also used for carrying out model training by using a preset multi-label multi-classification algorithm to obtain the class classification model.

3. An advertisement delivery method, comprising:

after receiving an access request of a user, a website acquires an identity identification number (CookieID) of the user and a corresponding user identity certificate (UID);

obtaining a classification result of the user by adopting the method according to any one of claims 1-2 according to the cookie ID and UID of the user;

taking the obtained category division result as a search keyword, and finding out advertisement content matched with the search keyword;

and playing the searched advertisement content in an advertisement playing column of the webpage returned to the user aiming at the access request.

4. A system for categorizing a user based on social behavior, comprising:

the user data acquisition module is used for acquiring and outputting user behavior data and user social data of users to be classified in a set period;

the attribute feature extraction module is used for extracting user behavior attribute features from the user behavior data output by the user data acquisition module and extracting user social attribute features from the user social data output by the user data acquisition module;

the user category dividing module is used for dividing the users into corresponding categories according to the user behavior attribute characteristics and the user social attribute characteristics extracted by the attribute characteristic extraction module; the user category classification module specifically comprises: a first user category dividing unit or a second user category dividing unit;

the system comprises a division model training module, a classification model selection module and a classification model selection module, wherein the division model training module is used for carrying out fusion and summarization to obtain the user characteristics of a training user after the user behavior attribute characteristics and the user social attribute characteristics of the training user are counted for each training user selected from users of a website; determining a behavior group to which the training user belongs according to the user behavior attribute characteristics of the training user, determining a social group to which the training user belongs according to the user social attribute characteristics of the training user, and taking a common group between the behavior group and the social group to which the training user belongs as the group to which the training user belongs; using the group of each training user and the user characteristics as training data, and performing model training by using a preset multi-label multi-classification algorithm to obtain a class classification model;

the first user category dividing unit specifically includes:

a behavior feature matching subunit, configured to match the user behavior attribute features extracted by the attribute feature extraction module with the user behavior attribute features recorded in a pre-established behavior feature rule base, and determine a behavior group to which the user belongs according to a matching result; the behavior feature rule base records user behavior attribute features of each behavior family for each behavior family;

the social characteristic matching subunit is used for matching the user social attribute characteristics extracted by the attribute characteristic extraction module with the user social attribute characteristics recorded in a pre-established social characteristic rule base and judging the social group to which the user belongs according to the matching result; the social feature rule base records user social attribute features of each social group;

the category dividing unit is used for receiving the behavior group to which the user belongs and output by the behavior feature matching unit and the social group to which the user belongs and output by the social feature matching unit; finding a common group between a behavior group and a social group to which the user belongs, and dividing the user into the found common group; and

the second user category dividing unit is used for fusing and summarizing the user behavior attribute features and the user social attribute features of the users extracted by the attribute feature extraction module to obtain the user features of the users; inputting the obtained user characteristics of the user into a category division model for category division; and classifying the users into corresponding categories through the category classification model.

5. An advertisement delivery system, comprising:

the user data acquisition module, the attribute feature extraction module, the user category classification module of claim 4;

the webpage request receiving module is used for acquiring the cookie ID and the corresponding UID of the user after receiving the access request of the user and sending the cookie ID and the corresponding UID to the user data acquiring module;

the advertisement query module is used for taking the classification result of the user output by the user classification module as a search keyword and searching the advertisement content matched with the search keyword;

the request processing module is used for returning the webpage to the user after inserting the advertisement content found by the advertisement inquiry module into the advertisement playing column of the webpage requested by the access request;

the user data acquisition module is specifically used for acquiring and outputting user behavior data and user social data of the user in a set period according to the cookie ID and the corresponding UID of the user.