CN121190155A

CN121190155A - Intelligent recommendation method, device, equipment and storage medium

Info

Publication number: CN121190155A
Application number: CN202511284000.3A
Authority: CN
Inventors: 魏涛; 许金秧; 邹建红; 陈奇
Original assignee: Shenzhen Yanxiang Stealth Technology Co ltd
Current assignee: Shenzhen Yanxiang Stealth Technology Co ltd
Priority date: 2025-09-09
Filing date: 2025-09-09
Publication date: 2025-12-23

Abstract

This application discloses an intelligent recommendation method, apparatus, device, and storage medium, relating to the field of product recommendation technology. The method includes: upon receiving a request from a target user, obtaining a target user profile of the target user; determining the target interest tag weights of the target user profile, wherein the target interest tag weights are determined based on basic interest tag weights, a time decay model, and a context enhancement factor, and are dynamically updated based on the target user's real-time behavioral data; and recommending corresponding products to the target user based on the target interest tag weights. This application can recommend more accurate products to users in real time.

Description

Intelligent recommendation method, device, equipment and storage medium

Technical Field

The application relates to the technical field of commodity recommendation, in particular to an intelligent recommendation method, device, equipment and storage medium.

Background

The commodity recommendation method mainly depends on user portraits to realize reasonable recommendation, but the traditional user portraits construction method based on batch processing generally carries out offline updating with the day level or even the week level as a period, and cannot timely reflect the dynamic change of the user interests, so that the recommendation result and the user demands can be deviated, and the shopping experience of the user is affected.

The foregoing is provided merely for the purpose of facilitating understanding of the technical solutions of the present application and is not intended to represent an admission that the foregoing is prior art.

Disclosure of Invention

The application mainly aims to provide an intelligent recommendation method, device, equipment and storage medium, and aims to solve the technical problem that a user interest change cannot be reflected timely to influence a recommendation result in a traditional user portrait construction method based on batch processing.

In order to achieve the above object, the present application provides an intelligent recommendation method, which includes:

when receiving request information of a target user, acquiring a target user portrait of the target user, determining target interest tag weights of the target user portrait, wherein the target interest tag weights are determined according to basic interest tag weights, a time decay model and a situation enhancement factor, the target interest tag weights are dynamically updated based on real-time behavior data of the target user, recommending corresponding commodities to the target user based on the target interest tag weights

In one embodiment, before the step of obtaining the target user representation of the target user, the method further includes:

acquiring user data of the target user, wherein the user data comprises behavior data, environment data and third party data;

Writing the user data into a local cache queue for asynchronous batch processing to obtain data to be processed, wherein the asynchronous batch processing comprises at least one of a dynamic buffer adjustment strategy, a priority strategy and an exception processing strategy;

After the data to be processed is issued to a preset message queue through Kafka, preprocessing the data to be processed by a stream processing engine to obtain target data, and constructing a target user portrait according to the target data, wherein the preprocessing comprises at least one of data cleaning, session cutting, feature extraction and anomaly detection.

In one embodiment, the step of determining the target interest tag weight for the target user representation comprises:

determining a behavior attenuation coefficient, behavior occurrence time and situation information according to the real-time behavior data;

Determining a time decay factor based on the time decay model, the behavior decay coefficient, and the behavior occurrence time;

determining a situation factor increasing rate and a situation factor reference value according to the situation information, and determining the situation enhancement factor according to the situation factor increasing rate and the situation factor reference value;

Real-time interest tag weights are determined based on the base interest tag weights, the time decay factor, and the context enhancement factor.

In an embodiment, the step of determining the context factor increase rate and the context factor reference value according to the context information, and determining the context enhancement factor according to the context factor increase rate and the context factor reference value comprises:

When the situation information is single enhancement information, determining the situation enhancement factors according to the situation factor increasing rate corresponding to the situation information and the situation factor reference value;

When the situation information is multi-enhancement information, determining a plurality of sub-situation enhancement factors according to the situation information, and determining the situation enhancement factors according to the sub-situation enhancement factors, wherein the sub-situation enhancement factors are determined according to a sub-situation factor increasing rate and a sub-situation factor reference value.

In one embodiment, the step of recommending the corresponding item to the target user based on the target interest tag weight includes:

determining a candidate commodity set according to the target interest weight label;

calculating the final score of each commodity in the candidate commodity set, and sorting the commodities in the candidate commodity set according to the final score to obtain a commodity sorting result;

Adjusting the commodity sequencing result based on a service technology adjustment strategy to obtain a target commodity sequencing result, wherein the service technology adjustment strategy comprises at least one of a diversity control technology, a freshness injection technology, a business rule engine technology, a compliance filtering technology and an exposure duplication removing technology;

And recommending corresponding commodities to the target user according to the target commodity sequencing result.

In one embodiment, the step of determining a candidate commodity set according to the target interest weight tag includes:

screening a first candidate commodity set from a commodity library according to the target interest weight label;

screening a second candidate commodity set from the commodity library according to the portrait vector of the target user portrait, and screening a third candidate commodity set from the commodity library according to the current hot information;

screening a fourth candidate commodity set from the commodity library according to the geographic position of the target user;

the candidate commodity set is determined based on the first candidate commodity set, the second candidate commodity set, the third candidate commodity set, and the fourth candidate commodity set.

In one embodiment, the step of calculating a final score for each item in the candidate set of items comprises:

Acquiring commodity characteristics of each commodity in the candidate commodity set, acquiring contextual characteristics of the target user, and determining user characteristics according to the target user portrait;

Inputting the commodity characteristics, the contextual characteristics and the user characteristics into a precision-arranging model to obtain an estimated click rate and an estimated conversion rate;

and determining the final score of each commodity in the candidate commodity set according to the estimated click rate and the estimated conversion rate based on a preset fusion strategy.

In addition, in order to achieve the above object, the present application also provides an intelligent recommendation device, which includes:

The acquisition module is used for acquiring a target user portrait of the target user when receiving request information of the target user;

The determining module is used for determining target interest tag weights of the target user portraits, wherein the target interest tag weights are determined according to basic interest tag weights, a time attenuation model and a situation enhancement factor, and the target interest tag weights are dynamically updated based on real-time behavior data of the target users;

and the recommending module is used for recommending corresponding commodities to the target user based on the target interest tag weight.

In addition, in order to achieve the above object, the application also proposes an intelligent recommendation device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the computer program being configured to implement the steps of the intelligent recommendation method as described above.

In addition, in order to achieve the above object, the present application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, implements the steps of the intelligent recommendation method as described above.

Furthermore, to achieve the above object, the present application provides a computer program product comprising a computer program which, when executed by a processor, implements the steps of the intelligent recommendation method as described above.

One or more technical schemes provided by the application have at least the following technical effects:

The intelligent recommendation method, the intelligent recommendation device, the intelligent recommendation equipment and the intelligent recommendation storage medium are characterized by acquiring a target user portrait of a target user when request information of the target user is received, determining target interest tag weights of the target user portrait, wherein the target interest tag weights are determined according to basic interest tag weights, a time attenuation model and situation enhancement factors, the target interest tag weights are dynamically updated based on real-time behavior data of the target user, and recommending corresponding commodities to the target user based on the target interest tag weights. Compared with the prior art, the method dynamically calculates the real-time interest tag weight by introducing the real-time behavior data stream and innovatively fusing the time attenuation model and the situation enhancement factor, realizes the user portrait construction mode transition from off-line batch processing to on-line real-time updating, can quickly sense and respond to the instant interest change of the user, remarkably improves the timeliness and accuracy of the recommendation result, effectively improves the user experience and improves the commercial conversion efficiency.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and together with the description, serve to explain the principles of the application.

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are required to be used in the description of the embodiments or the prior art will be briefly described below, and it will be obvious to those skilled in the art that other drawings can be obtained from these drawings without inventive effort.

FIG. 1 is a schematic flow chart of an intelligent recommendation method according to an embodiment of the present application;

FIG. 2 is a schematic flow chart of an intelligent recommendation method according to an embodiment of the present application;

FIG. 3 is a diagram of a user portrait construction technology provided in a first embodiment of the intelligent recommendation method of the present application;

FIG. 4 is a framework diagram of a weight update core algorithm provided in a second embodiment of an intelligent recommendation method according to the present application;

FIG. 5 is a schematic diagram of a multi-path recall technique according to a second embodiment of the intelligent recommendation method of the present application;

FIG. 6 is a schematic diagram of a ranking service technique according to a second embodiment of the intelligent recommendation method of the present application;

FIG. 7 is a schematic block diagram of an intelligent recommendation device according to an embodiment of the present application;

fig. 8 is a schematic device structure diagram of a hardware operating environment related to an intelligent recommendation method in an embodiment of the present application.

The achievement of the objects, functional features and advantages of the present application will be further described with reference to the accompanying drawings, in conjunction with the embodiments.

Detailed Description

It should be understood that the specific embodiments described herein are merely illustrative of the technical solution of the present application and are not intended to limit the present application.

For a better understanding of the technical solution of the present application, the following detailed description will be given with reference to the drawings and the specific embodiments.

The method comprises the main steps of obtaining a target user portrait of a target user when request information of the target user is received, determining target interest tag weights of the target user portrait, wherein the target interest tag weights are determined according to basic interest tag weights, a time attenuation model and a situation enhancement factor, the target interest tag weights are dynamically updated based on real-time behavior data of the target user, and recommending corresponding commodities to the target user based on the target interest tag weights.

According to the embodiment, when the request information of the target user is received, the target user portrait of the target user is obtained, the target interest tag weight of the target user portrait is determined, wherein the target interest tag weight is determined according to the basic interest tag weight, the time attenuation model and the situation enhancement factor, the target interest tag weight is dynamically updated based on the real-time behavior data of the target user, and corresponding commodities are recommended to the target user based on the target interest tag weight. Compared with the prior art, the method dynamically calculates the real-time interest tag weight by introducing the real-time behavior data stream and innovatively fusing the time attenuation model and the situation enhancement factor, realizes the user portrait construction mode transition from off-line batch processing to on-line real-time updating, can quickly sense and respond to the instant interest change of the user, remarkably improves the timeliness and accuracy of the recommendation result, effectively improves the user experience and improves the commercial conversion efficiency.

It should be noted that, the execution body of the embodiment may be a computing service device with functions of data processing, network communication and program running, such as a tablet computer, a personal computer, a mobile phone, or an electronic device, an intelligent recommendation device, or the like, which can implement the above functions. The present embodiment and the following embodiments will be described below by taking an intelligent recommendation device as an example.

Based on this, an embodiment of the present application provides an intelligent recommendation method, and referring to fig. 1, fig. 1 is a schematic flow chart of a first embodiment of the intelligent recommendation method of the present application.

In this embodiment, the intelligent recommendation method includes steps S10 to S30:

Step S10, when request information of a target user is received, a target user portrait of the target user is obtained;

it should be noted that, according to the user identifier (such as userID) carried in the request information, the user portrait corresponding to the user can be quickly retrieved and loaded from the user portrait database.

In a possible implementation manner, before the step of obtaining the target user portrait of the target user, the method further comprises the steps of obtaining user data of the target user, wherein the user data comprise behavior data, environment data and third party data, writing the user data into a local cache queue for asynchronous batch processing to obtain data to be processed, the asynchronous batch processing comprises at least one of a dynamic buffer adjustment strategy, a priority strategy and an exception handling strategy, after the data to be processed is issued to a preset message queue through Kafka, preprocessing the data to be processed by a stream processing engine to obtain target data, and constructing the target user portrait according to the target data, wherein the preprocessing comprises at least one of data cleaning, session cutting, feature extraction and exception detection.

It should be noted that, in addition to acquiring user behavior data, environmental context may be acquired, where the environmental context may be information such as GPS location, device type, time, and the like, and third party data may be social media, weather, and news events.

It should be noted that, the core function of setting the local caching strategy is to cut peaks and fill valleys, cope with instantaneous flow peaks, avoid direct impact on Kafka, compress in batch, combine multiple data into a single network request, reduce bandwidth consumption, and provide temporary storage when network jitter or Kafka is unavailable.

It can be understood that the dynamic buffer adjustment policy can automatically adjust the batch processing size and the sending interval according to the system load (CPU/memory), the priority policy can distinguish the critical event (such as successful payment) from the common event (such as page browsing), and ensure the priority sending of the high priority data, for example, the dual-queue design concept can be set, the high priority queue and the common queue can be used for realizing the priority sending of the critical service event (such as payment) through the dual queue, so as to avoid the delay of important data due to the queue accumulation, the abnormal processing policy is used for summarizing the process of sending the data to Kafka, a mechanism of failed retry exists, the number of failed times can be set in a self-defined manner, but the excessively large data is not recommended to be set, the default is 3 times, the data can be written into the local file after the 3 times of retry still fails, and the timing task can be used for processing the failed data within the set time.

It should be noted that, the data to be processed can be sent to the Kafka producer through the data compression and encryption algorithm, the compression algorithm (such as Snappy) is used for saving the broadband cost, and the encryption scheme (such as AES-GCM) is used for guaranteeing the data transmission safety. Kafka producer receives encrypted data, uses APACHE KAFKA partition strategy, has the advantages of high throughput, realizes million-level TPS and guarantees data persistence. The single cluster actual measurement supports 100 ten thousand TPS, and the linear expansion can be realized by adding partitions and consumers. Low latency and real-time guarantee, data generation to consumable latency <100ms. The data reliability design, the multi-copy mechanism and the ISR synchronization guarantee the consistency of the data, and the message retention strategy supports the retention of the original data according to time/size (such as 7 days). Ecological seamless integration, fink directly consumes Kafka.

It should be noted that, after the data to be processed is issued to a preset message queue through Kafka, the stream processing engine performs preprocessing on the data to be processed. The Flink (i.e. the stream processing engine) can directly consume the data of Kafka, a high throughput low delay channel of Kafka- > Flink- > downstream system is established, and the stream ETL engine is used for data real-time conversion, so that complex event processing can be dealt with.

It should be noted that, as shown in fig. 2, preprocessing performed on data to be processed by the link (i.e., the stream processing engine) includes data cleansing and normalization, session cutting, real-time feature extraction, and abnormal behavior detection. The method comprises the steps of carrying out data check and data conversion according to standardization on data to be processed, wherein the data conversion comprises JSON analysis, field check, standardization conversion, addition of processing time stamp, error data processing and the like, carrying out grouping on the data to be processed mainly for session cutting, collecting all event data in a session, calculating session characteristics, outputting complete session, carrying out counting statistics on one counting feature of user behaviors mainly for real-time feature extraction, generating feature vectors, adding real-time features, carrying out calculation on click frequency of a user mainly for abnormal behavior detection, and if the click frequency is larger than a specified value (such as 100 times/min), regarding the abnormal behavior data as abnormal, filtering the abnormal behavior data and recording the abnormal behavior data in an abnormal log. And finally, constructing the target user portrait according to the target data obtained by preprocessing.

Step S20, determining target interest tag weights of the target user portraits, wherein the target interest tag weights are determined according to basic interest tag weights, a time attenuation model and a situation enhancement factor, and the target interest tag weights are dynamically updated based on real-time behavior data of the target users;

It should be noted that the basic interest tag weight refers to the interest tag weight before the real-time behavior data is updated, the target interest tag weight refers to the interest tag weight after the real-time behavior data is updated, the interest tag weight is a core quantization index in the user portrait and can be used for representing the interest degree of the user on a certain type of commodity, and the user's behavior often has commodity information (such as Zhang San searching mobile phone and clicking), at this time, the user is marked with a label (such as mobile phone) of commodity information attribute according to the commodity information associated with the user's behavior, and then the interest tag weight of the user on the commodity is updated according to the user's behavior.

It should be noted that, the real-time behavior data refers to the access behavior of the user on the system page, where the access behavior includes a purchase behavior, a payment behavior, a purchasing behavior, a collection behavior, a searching behavior, a sharing behavior, a clicking behavior, and a browsing behavior, and the real-time behavior data further includes some time information and context information (such as what time period and what position the access behavior of the user occurs), and the target interest tag weight in the target user image may be dynamically updated according to the real-time behavior data.

In a specific implementation, the collected real-time behavior data, environment context and third party data can be checked, including integrity check, validity check and consistency check. Specifically, for integrity check, check whether a necessary field exists (e.g., eventId, userId, timetable), whether a nested field structure conforms to a protocol (e.g., whether subfields in properties are complete), etc., for validity check, check field format (e.g., whether a timestamp is 13 bits, IP address format), numerical range (e.g., commodity price cannot be negative), enumerate value validity (e.g., whether eventType is in an allowed list), etc., for consistency check, check cross-field logic (e.g., that the time of a following order cannot be earlier than the time of purchase), etc.

In one possible implementation, the step of determining the target interest tag weight of the target user portrait includes determining a behavior decay coefficient, a behavior occurrence time, and context information from the real-time behavior data, determining a time decay factor based on the time decay model, the behavior decay coefficient, and the behavior occurrence time, determining a context factor increase rate and a context factor reference value from the context information, and determining the context enhancement factor based on the context factor increase rate and the context factor reference value, and determining a real-time interest tag weight based on the base interest tag weight, the time decay factor, and the context enhancement factor.

It should be noted that, for the basic interest tag weight in the target user portrait, the initial basic interest tag weight may be determined according to the behavior type, for example, purchase behavior is 0.85, payment behavior is 0.80, purchasing behavior is 0.70, collection behavior is 0.65, search behavior is 0.60, sharing behavior is 0.40, clicking behavior is 0.35, and browsing behavior is 0.25. And taking the real-time interest tag weight obtained based on the real-time behavior data update every time as the basic interest tag weight of the new real-time interest tag weight obtained by the next real-time dynamic update.

It should be noted that, the calculation formula of the time attenuation factor is as follows:

w_t＝e^-λt

Where λ represents the behavior decay coefficient, t represents the number of minutes until the behavior event occurs (i.e., the behavior occurrence time), and different behavior decay coefficients may be configured according to different behavior types. Exemplary purchase behavior is 0.0015, payment behavior is 0.0015, additional purchase behavior is 0.002, collection behavior is 0.002, search behavior is 0.003, sharing behavior is 0.005, clicking behavior is 0.005, and browsing behavior is 0.004.

It should be noted that, the calculation formula of the context enhancement factor is as follows:

w(k)=(k+1)v

Where k represents a context enhancement factor increase rate, v represents a context factor reference value, which may be set according to actual conditions, for example, to 1.

It should be noted that, as shown in fig. 3, the formula for calculating the real-time interest tag weight by combining the basic interest tag weight and the time attenuation factor by the context enhancement factor is as follows:

W_n＝W₁*w_t*w(k)

Where W (k) represents a context enhancement factor, W _t represents a time decay factor, and W ₁ represents a base interest tag weight.

In one possible implementation, the step of determining the context factor increasing rate and the context factor reference value according to the context information and determining the context enhancement factor according to the context factor increasing rate and the context factor reference value includes determining the context enhancement factor according to the context factor increasing rate corresponding to the context information and the context factor reference value when the context information is single enhancement information, determining a plurality of sub-context enhancement factors according to the context information when the context information is multi-enhancement information, and determining the context enhancement factor according to the sub-context enhancement factors, wherein the sub-context enhancement factor is determined according to the sub-context factor increasing rate and the sub-context factor reference value.

The context information includes at least one of time zone enhancement information, location enhancement information, device type enhancement information, and promotional activity enhancement information. When the context information only relates to one type of enhancement information, the context enhancement factor increasing rate and the context factor reference value corresponding to the enhancement information can be directly obtained to determine the context enhancement factor. Specifically, for a period enhancement, such as a late peak (18-22) period, the context factor increase rate is 30%, and the context enhancement factor is changed to 1.3 at this time. For location augmentation, i.e. a specific scene weighting, such as a mall environment, the context factor increase rate is 40% and the context augmentation factor is changed to 1.4 at this time. For device type enhancement, such as mobile-side weighting, the context factor increase rate is 20%, and the context enhancement factor is changed to 1.2 at this time. For the promotion enhancement, if the scene of the promotion is, the increase rate of the situation factor is 50%, and the situation enhancement factor is changed to 1.5 at this time.

The context information includes at least one of time zone enhancement information, location enhancement information, device type enhancement information, and promotional activity enhancement information. When the context information relates to various enhancement information, the sub-context enhancement factor increasing rate and the sub-context factor reference value corresponding to each enhancement information (namely sub-context enhancement information) can be respectively obtained to determine the sub-context enhancement factors, and then the context enhancement factors are determined according to all the sub-context enhancement factors. Specifically, for example, if the context information relates to time period enhancement and location enhancement, then the sub-context enhancement factor corresponding to the time period enhancement (the sub-context factor increase rate is 30% in the late peak time period, the sub-context enhancement factor is changed to 1.3 at this time), the sub-context enhancement factor corresponding to the location enhancement (such as the market environment, the sub-context factor increase rate is 40% and the sub-context enhancement factor is changed to 1.4 at this time) may be calculated, and finally the final context enhancement factor is determined according to all the sub-context enhancement factors, which is calculated as follows:

w(k)=w(k₁)*w(k₂)*...*w(k_n),w(k_n)=(k+1)v

Where w (k) represents a context enhancement factor and w (k _n) represents a sub-context enhancement factor.

And step S30, recommending corresponding commodities to the target user based on the target interest tag weight.

In a specific implementation, candidate commodity sets which are relatively related to the interests of the user can be rapidly screened out from a mass commodity library according to the weights of the target interest labels, specifically, core interest labels (such as Top-N labels with highest weights) of the user are selected according to weight sorting, the commodity IDs related to the labels are rapidly searched out by utilizing an efficient inverted index technology to form commodity candidate sets, and corresponding commodities in the commodity candidate sets are recommended to the target user. And a multi-path recall strategy can be adopted to expand the commodity candidate set, for example, the commodity which is liked by a user group which is similar to the target user recently is recalled by combining collaborative filtering, or the high-heat commodity is recalled from a popular commodity library, so that individuation and popularity of the recommendation result are considered.

The method comprises the steps of obtaining a target user portrait of a target user when request information of the target user is received, determining target interest tag weights of the target user portrait, wherein the target interest tag weights are determined according to basic interest tag weights, a time attenuation model and a situation enhancement factor, the target interest tag weights are dynamically updated based on real-time behavior data of the target user, and recommending corresponding commodities to the target user based on the target interest tag weights. Compared with the prior art, the method dynamically calculates the real-time interest tag weight by introducing the real-time behavior data stream and innovatively fusing the time attenuation model and the situation enhancement factor, realizes the user portrait construction mode transition from off-line batch processing to on-line real-time updating, can quickly sense and respond to the instant interest change of the user, remarkably improves the timeliness and accuracy of the recommendation result, effectively improves the user experience and improves the commercial conversion efficiency.

In the second embodiment of the present application, the same or similar content as that of the first embodiment may be referred to the description above, and will not be repeated. On this basis, referring to fig. 4, step S30 further includes steps S301 to S304:

Step S301, determining a candidate commodity set according to the target interest weight label;

In a specific implementation, the Top N interest tags (e.g., top 20) with the highest weight may be extracted from the target user representation, which is used as a core representation of the current interest of the user. And then using the pre-constructed inverted index (Inverted Index) data structure and taking the high-weight labels as query keys to quickly search all the commodity ID sets marked with the same or similar labels. In a possible implementation manner, the step of determining the candidate commodity set according to the target interest weight label comprises the steps of screening a first candidate commodity set from a commodity library according to the target interest weight label, screening a second candidate commodity set from the commodity library according to the portrait vector of the target user portrait, screening a third candidate commodity set from the commodity library according to the current trending information, screening a fourth candidate commodity set from the commodity library according to the geographic position of the target user, and determining the candidate commodity set based on the first candidate commodity set, the second candidate commodity set, the third candidate commodity set and the fourth candidate commodity set.

It should be noted that, in order to improve the recall coverage rate and precision, as shown in fig. 5, a multi-path recall strategy is generally adopted, for example, a combination behavior collaborative filtering mode, a popular commodity mode and other recall modes are combined, and the results returned by each path are combined and de-duplicated, so that a candidate commodity set with controllable scale and higher correlation is finally formed and is sent to a downstream sequencing module.

Step S302, calculating the final score of each commodity in the candidate commodity set, and sorting the commodities in the candidate commodity set according to the final score to obtain a commodity sorting result;

In a possible implementation manner, the step of calculating the final score of each commodity in the candidate commodity set includes the steps of obtaining commodity characteristics of each commodity in the candidate commodity set, obtaining context characteristics of the target user, determining user characteristics according to the target user portrait, inputting the commodity characteristics, the context characteristics and the user characteristics into a precision ranking model to obtain an estimated click rate and an estimated conversion rate, and determining the final score of each commodity in the candidate commodity set according to the estimated click rate and the estimated conversion rate based on a preset fusion strategy.

It should be noted that, the user features mainly include user portraits, user history CTR (i.e. the number of times the user clicks the recommended commodity/the number of times the commodity is pushed to the user, for example, the system recommends 1000 commodities to the user a, the user a clicks 200, the history CTR of the user is 20%), 7 days of liveness, consumption capability, etc., the commodity features mainly include category labels, price segments, commodity history CTR (i.e. the number of times the commodity=clicked/the number of times the commodity is displayed, for example, 100 ten thousand times after the commodity a is online, 1 ten thousand times after the commodity a is clicked, its commodity history CTR is 1%), inventory status, etc., and the context features mainly include geographic location, time stamp, device type, network environment, etc. The estimated click rate refers to the probability of clicking a commodity by a user, and the estimated conversion rate refers to the probability of final purchase after clicking the commodity by the user.

In a specific implementation, the fine-grained model includes a deep interest Network (i.e., DIN, deep Interest Network) and a Multi-gate hybrid expert Network (i.e., MMOE, multi-gate mix-of-expertise). The deep interest network is mainly used for deep digging historical behaviors of users, finding out relevant information (namely user-commodity interest vectors) of the historical behaviors and current commodities, splicing the user-commodity interest vectors with other characteristics (such as commodity characteristics and contextual characteristics), inputting the spliced user-commodity interest vectors into the multi-gate hybrid expert network, and predicting estimated click rate and estimated conversion rate.

In a specific implementation, a calculation formula for calculating a final score of the commodity according to the estimated click rate and the estimated conversion rate is as follows:

FinalScore=0.7*pCTR+0.3*pCVR

where pCTR represents the estimated click rate and pCVR represents the estimated conversion rate.

Step S303, adjusting the commodity sorting result based on a service technology adjustment strategy to obtain a target commodity sorting result, wherein the service technology adjustment strategy comprises at least one of a diversity control technology, a freshness injection technology, a business rule engine technology, a compliance filtering technology and an exposure deduplication technology;

It should be noted that, as shown in fig. 6, the data output through the ordering layer needs to be readjusted based on the service technology adjustment policy. Specifically, for diversity control technology, mainly used solves the similar mesh heap problem of sequencing result, can break up data category through the cluster deduplication, avoid the information cocoon room, promote exploratory, specifically confirm commodity category according to commodity sequencing result to select TOP1 commodity from different commodity categories and reorder. The key technology for resisting information cocoons and solving the Martai effect is characterized in that the key design idea is that new contents which are not contacted by a user are introduced strategically on the premise of guaranteeing recommendation correlation, the injection of new commodities can be realized by setting the proportion of dynamic injection, specifically, commodity categories can be determined according to commodity sorting results, other commodity categories related to the commodity categories can be determined, and then commodities of the other commodity categories are added into commodity sorting results. For the business rule engine technology, a core component mainly used for responding to operation requirements is to manually configure business rules to meet the operation activity requirements, specifically, for example, new commodities are added into commodity sorting results according to the business rules configured during the period of 618, and further, for example, new commodities are added into commodity sorting results according to VIP business rules. For the compliance filtering technology, the method is an important defense line for legal risk prevention and control, a sensitive/prohibit selling word stock is required to be accessed, sensitive word detection is carried out on commodity information in commodity sequencing results, qualification is checked, and whether the license is complete is checked. For the exposure de-duplication technology, the user is prevented from repeatedly seeing the same content in a short time, specifically, the latest commodity exposure record of the user can be obtained, and then the exposed commodities in the commodity sorting result are filtered out according to the commodity exposure record.

And step S304, recommending corresponding commodities to the target user according to the target commodity sequencing result.

It should be noted that, according to the sequence of the commodities in the target commodity sorting result, a plurality of commodities with highest rank can be sequentially displayed to the target user. Such a presentation is typically embodied in a stream of information on the application front page, a "guess you like" module for the item detail page, or a personalized push message, among other scenarios, that will most likely attract the attention of the user and create an interactive preferential exposure of the item. Through the accurate recommendation based on the ordering, the calculated user interests and commodity values can be converted into actual business effects to the greatest extent, the click rate, the purchase conversion rate and the overall user experience of the user are effectively improved, and finally the closed loop from data perception to commercial value realization is completed.

The method comprises the steps of determining a candidate commodity set according to the target interest weight label, calculating the final score of each commodity in the candidate commodity set, sorting the commodities in the candidate commodity set according to the final score to obtain a commodity sorting result, adjusting the commodity sorting result based on a service technology adjustment strategy to obtain a target commodity sorting result, wherein the service technology adjustment strategy comprises at least one of a diversity control technology, a freshness injection technology, a business rule engine technology, a compliance filtering technology and an exposure duplication removing technology, and recommending corresponding commodities to a target user according to the target commodity sorting result. Through the method, accurate commodity preliminary screening and sorting can be achieved based on real-time interests of users, and the problems that pure algorithm recommendation is likely to cause homogenization, old content, non-conforming to business strategies or poor user experience and the like are effectively solved by comprehensively applying various adjustment technologies, so that diversity, novelty, compliance and business suitability of recommendation results are obviously improved while recommendation relevance is guaranteed, and finally a more balanced, robust and user-satisfied personalized recommendation scheme is formed.

It should be noted that the foregoing examples are only for understanding the present application, and do not constitute a limitation of the intelligent recommendation method of the present application, and many simple changes based on this technical concept are all within the scope of the present application.

The present application also provides an intelligent recommendation device, referring to fig. 7, the intelligent recommendation device includes:

an acquisition module 10, configured to acquire a target user portrait of a target user when receiving request information of the target user;

A determining module 20, configured to determine a target interest tag weight of the target user portrait, where the target interest tag weight is determined according to a base interest tag weight, a time decay model, and a context enhancement factor, and the target interest tag weight is dynamically updated based on real-time behavior data of the target user;

and the recommending module 30 is used for recommending corresponding commodities to the target user based on the target interest tag weight.

The intelligent recommendation device provided by the application can solve the technical problem that the traditional user portrait construction method based on batch processing cannot timely reflect the user interest change to influence the recommendation result by adopting the intelligent recommendation method in the embodiment. Compared with the prior art, the intelligent recommendation device has the same beneficial effects as the intelligent recommendation method provided by the embodiment, and other technical features in the intelligent recommendation device are the same as the disclosed features of the method of the embodiment, and are not repeated herein.

The application provides intelligent recommendation equipment, which comprises at least one processor and a memory in communication connection with the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the intelligent recommendation method in the first embodiment.

Referring now to FIG. 8, a schematic diagram of an intelligent recommendation device suitable for use in implementing embodiments of the present application is shown. The intelligent recommendation device in the embodiment of the present application may include, but is not limited to, mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (Personal DIGITAL ASSISTANT: personal digital assistants), PADs (Portable Application Description: tablet computers), PMPs (Portable MEDIA PLAYER: portable multimedia players), car terminals (e.g., car navigation terminals), and the like, and fixed terminals such as digital TVs, desktop computers, and the like. The intelligent recommendation device illustrated in fig. 8 is merely an example, and should not be construed as limiting the functionality and scope of use of embodiments of the present application.

As shown in fig. 8, the intelligent recommendation apparatus may include a processing device 1001 (e.g., a central processing unit, a graphic processor, etc.) that may perform various appropriate actions and processes according to programs stored in a read-only memory 1002 or programs loaded from a storage device 1003 into a random access memory 1004. In the random access memory 1004, various programs and data required for the intelligent recommendation device operation are also stored. The processing device 1001, the read only memory 1002, and the random access memory 1004 are connected to each other by a bus 1005. An input/output interface 1006 is also connected to the bus. In general, a system including an input device 1007 such as a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc., an output device 1008 including a Liquid crystal display (LCD: liquid CRYSTAL DISPLAY), a speaker, a vibrator, etc., a storage device 1003 including a magnetic tape, a hard disk, etc., and a communication device 1009 may be connected to the input/output interface 1006. The communicator 1009 may allow the intelligent recommendation device to communicate wirelessly or by wire with other devices to exchange data. While intelligent recommendation devices having various systems are shown in the figures, it should be understood that not all of the illustrated systems are required to be implemented or provided. More or fewer systems may alternatively be implemented or provided.

In particular, according to embodiments of the present disclosure, the processes described above with reference to flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising program code for performing the method shown in the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network via a communication device, or installed from the storage device 1003, or installed from the read only memory 1002. The above-described functions defined in the method of the disclosed embodiment of the application are performed when the computer program is executed by the processing device 1001.

The intelligent recommendation equipment provided by the application can solve the technical problem that the traditional user portrait construction method based on batch processing cannot timely reflect the user interest change to influence the recommendation result by adopting the intelligent recommendation method in the embodiment. Compared with the prior art, the intelligent recommendation device provided by the application has the same beneficial effects as the intelligent recommendation method provided by the embodiment, and other technical features in the intelligent recommendation device are the same as the features disclosed by the method of the previous embodiment, and are not repeated herein.

It is to be understood that portions of the present disclosure may be implemented in hardware, software, firmware, or a combination thereof. In the description of the above embodiments, particular features, structures, materials, or characteristics may be combined in any suitable manner in any one or more embodiments or examples.

The foregoing is merely illustrative of the present application, and the present application is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

The present application provides a computer readable storage medium having computer readable program instructions (i.e., a computer program) stored thereon for performing the intelligent recommendation method in the above-described embodiments.

The computer readable storage medium provided by the present application may be, for example, a USB flash disk, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system or device, or a combination of any of the foregoing. More specific examples of a computer-readable storage medium may include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access Memory (RAM: random Access Memory), a Read-Only Memory (ROM), an erasable programmable Read-Only Memory (EPROM: erasable Programmable Read Only Memory or flash Memory), an optical fiber, a portable compact disc Read-Only Memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In this embodiment, the computer-readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system or device. Program code embodied on a computer readable storage medium may be transmitted using any appropriate medium, including but not limited to electrical wiring, fiber optic cable, RF (Radio Frequency) and the like, or any suitable combination of the foregoing.

The computer readable storage medium may be included in the intelligent recommendation device or may exist alone without being assembled into the intelligent recommendation device.

The computer readable storage medium carries one or more programs, when the one or more programs are executed by the intelligent recommendation device, the intelligent recommendation device obtains a target user portrait of a target user when request information of the target user is received, and determines target interest tag weights of the target user portrait, wherein the target interest tag weights are determined according to basic interest tag weights, a time attenuation model and a situation enhancement factor, the target interest tag weights are dynamically updated based on real-time behavior data of the target user, and corresponding commodities are recommended to the target user based on the target interest tag weights.

Computer program code for carrying out operations of the present application may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, smalltalk, C ++ and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of remote computers, the remote computer may be connected to the user's computer through any kind of network, including a local area network (LAN: localArea Network) or a wide area network (WAN: wide Area Network), or may be connected to an external computer (for example, through the Internet using an Internet service provider).

The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The modules involved in the embodiments of the present application may be implemented in software or in hardware. Wherein the name of the module does not constitute a limitation of the unit itself in some cases.

The readable storage medium provided by the application is a computer readable storage medium, and the computer readable storage medium is stored with computer readable program instructions (namely computer program) for executing the intelligent recommendation method, so that the technical problem that the recommendation result is influenced because the user interest change cannot be reflected in time by the traditional batch-based user portrait construction method can be solved. Compared with the prior art, the beneficial effects of the computer readable storage medium provided by the application are the same as those of the intelligent recommendation method provided by the embodiment, and are not repeated here.

The application also provides a computer program product comprising a computer program which, when executed by a processor, implements the steps of the intelligent recommendation method as described above.

The computer program product provided by the application can solve the technical problem that the recommendation result is affected because the user interest change cannot be reflected in time by the traditional user portrait construction method based on batch processing. Compared with the prior art, the beneficial effects of the computer program product provided by the application are the same as the beneficial effects of the intelligent recommendation method provided by the embodiment, and are not repeated here.

The foregoing description is only a partial embodiment of the present application, and is not intended to limit the scope of the present application, and all the equivalent structural changes made by the description and the accompanying drawings under the technical concept of the present application, or the direct/indirect application in other related technical fields are included in the scope of the present application.

Claims

1. An intelligent recommendation method is characterized by comprising the following steps:

when receiving request information of a target user, acquiring a target user portrait of the target user;

determining a target interest tag weight of the target user portrait, wherein the target interest tag weight is determined according to a basic interest tag weight, a time decay model and a situation enhancement factor, and the target interest tag weight is dynamically updated based on real-time behavior data of the target user;

And recommending corresponding commodities to the target user based on the target interest tag weight.

2. The intelligent recommendation method of claim 1, wherein prior to the step of obtaining the target user representation of the target user, further comprising:

3. The intelligent recommendation method of claim 1, wherein the step of determining a target interest tag weight for the target user representation comprises:

4. The method of claim 3, wherein the steps of determining a context factor increase rate and a context factor reference value from the context information and determining the context enhancement factor from the context factor increase rate and the context factor reference value comprise:

5. The intelligent recommendation method of claim 1, wherein the step of recommending respective items to the target user based on the target interest tag weights comprises:

6. The intelligent recommendation method of claim 5, wherein the step of determining a candidate merchandise set based on the target interest weight tag comprises:

7. The intelligent recommendation method of claim 5, wherein the step of calculating a final score for each item in the set of candidate items comprises:

8. An intelligent recommendation device, characterized in that the intelligent recommendation device comprises:

9. An intelligent recommendation device, characterized in that the device comprises a memory, a processor and a computer program stored on the memory and executable on the processor, the computer program being configured to implement the steps of the intelligent recommendation method according to any one of claims 1 to 7.

10. A storage medium, characterized in that the storage medium is a computer-readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, implements the steps of the intelligent recommendation method according to any of claims 1 to 7.