Object recommendation method and device, electronic equipment and readable storage medium

Abstract

The invention relates to the technical field of object recommendation, and provides an object recommendation method and device, electronic equipment and a readable storage medium. The method comprises the following steps: performing session division processing on a user historical behavior feature vector to obtain at least one user historical session vector; processing each user historical session vector and the attribute vector corresponding to the to-be-recommended object based on an attention mechanism to obtain each self-attention vector; performing gating mechanism processing on each user historical session vector and each self-attention vector to obtain each gating vector; performing fusion processing on each gating vector to obtain a user interest vector; determining a recommendation score corresponding to the to-be-recommended object according to the user attribute vector, the user interest vector and an attribute vector corresponding to the to-be-recommended object; and determining the target object data displayed to the user based on the preset recommendation score threshold value and the recommendation score, thereby improving the accuracy of user interest prediction and enhancing the capture ability of dynamic interest changes of the user.

Description

Technical Field

[0001] The present disclosure relates to the technical field of object recommendation, and in particular to an object recommendation method, device, electronic device and readable storage medium. Background Art

[0002] Generally speaking, mining user interests from large amounts of data is the core of the recommendation system field. User behaviors are usually modeled based on user behavior sequence models through attention mechanisms. However, traditional models mainly focus on individual behaviors, fail to fully utilize the temporal information in user behavior sequences, and fail to capture user interest changes and dependencies. When processing user interests, existing technologies generally treat user behaviors as isolated events, resulting in the neglect of user behavior patterns in specific sessions, which limits the model's ability to capture dynamic changes in user interests.

[0003] It can be seen that the existing technology has the problem of low prediction accuracy of user interest evolution and poor accuracy of recommendation results due to failure to pay attention to the continuous evolution characteristics of user behavior sessions. Summary of the invention

[0004] In view of this, the embodiments of the present disclosure provide an object recommendation method, device, electronic device and readable storage medium to solve the problem in the prior art that the continuous evolution characteristics of user behavior sessions are not paid attention to, resulting in low prediction accuracy of user interest evolution and poor accuracy of recommendation results.

[0005] According to a first aspect of an embodiment of the present disclosure, there is provided an object recommendation method, comprising: obtaining a user attribute vector, a user historical behavior feature vector, and an attribute vector corresponding to an object to be recommended, and inputting the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended into a trained object recommendation model; performing session division processing on the user historical behavior feature vector to obtain at least one user historical session vector; performing a gating mechanism processing on each user historical session vector and the attribute vector corresponding to the object to be recommended to obtain a self-attention vector corresponding to each user historical session vector; performing a gating mechanism processing on each user historical session vector and the self-attention vector corresponding to each user historical session vector to obtain a gating vector corresponding to each user historical session vector; performing a fusion processing on the gating vectors corresponding to each user historical session vector to obtain a user interest vector; determining a recommendation score corresponding to the object to be recommended according to the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended; determining target object data to be displayed to the user based on a preset recommendation score threshold and the recommendation score corresponding to the object to be recommended, and sending the target object data to a target terminal device so that the target object data is displayed on a current graphical user interface of the target terminal device.

[0006] According to a second aspect of the embodiments of the present disclosure, an object recommendation device is provided, including: an acquisition module, which is used to acquire a user attribute vector, a user historical behavior feature vector, and an attribute vector corresponding to a to-be-recommended object, and input the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the to-be-recommended object into a trained object recommendation model; a first processing module, which is used to perform session division processing on the user historical behavior feature vector to obtain at least one user historical session vector; a second processing module, which is used to process each user historical session vector and the attribute vector corresponding to the to-be-recommended object based on an attention mechanism to obtain a self-attention vector corresponding to each user historical session vector; and a third processing module, which is used to perform session division processing on each user historical behavior feature vector to obtain a self-attention vector corresponding to each user historical session vector. The historical conversation vector and the self-attention vector corresponding to each user's historical conversation vector are processed by a gating mechanism to obtain the gating vector corresponding to each user's historical conversation vector; the fourth processing module is used to fuse the gating vectors corresponding to each user's historical conversation vector to obtain the user interest vector; the determination module is used to determine the recommendation score corresponding to the object to be recommended according to the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended; the fifth processing module is used to determine the target object data to be displayed to the user based on a preset recommendation score threshold and the recommendation score corresponding to the object to be recommended, and send the target object data to the target terminal device, so that the target object data is displayed on the current graphical user interface of the target terminal device.

[0007] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0008] According to a fourth aspect of an embodiment of the present disclosure, a readable storage medium is provided, which stores a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0009] Compared with the prior art, the embodiments of the present disclosure have the following beneficial effects: by obtaining a user attribute vector, a user historical behavior feature vector, and an attribute vector corresponding to an object to be recommended that are authorized by the user, and inputting the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended into a pre-trained object recommendation model for subsequent processing, the user historical behavior feature vector is divided into sessions, and the user historical behavior feature vector is divided into at least one user historical session vector, and then each user historical session vector and the attribute vector corresponding to the object to be recommended can be processed through an attention mechanism, and a self-attention vector corresponding to each user historical session vector can be obtained after processing through the attention mechanism, and then each user historical session vector and the self-attention vector corresponding to each user historical session vector can be processed through a gating mechanism to obtain a gating vector corresponding to each user historical session vector, and the gating vectors corresponding to each user historical session vector can be fused to obtain a user interest vector, which can be used to characterize the user's interest in a certain area. The recommendation score corresponding to the object to be recommended can be calculated based on the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended. The recommendation score can be used to characterize the user's interest in the object to be recommended, and the recommendation score corresponding to the object to be recommended can be judged and screened based on a preset recommendation score threshold. If the recommendation score corresponding to the object to be recommended is greater than or equal to the preset recommendation score threshold, the object to be recommended can be determined as the target object data that can be displayed to the user, and the target object data can be sent to the target terminal device used by the user, for displaying the target object data on the current graphical user interface of the target terminal device. In this way, the personalization of the recommendation is improved by considering the user attribute vector and the user's historical behavior feature vector. Through processing based on the attention mechanism, the object recommendation model's ability to capture the user's dynamic interests is improved. A gating mechanism is introduced to process the user's historical session vector and the corresponding self-attention vector, which reduces noise interference and enhances the scalability and flexibility of the object recommendation model. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0011] Figure 1 is a schematic diagram of an application scenario of an embodiment of the present disclosure;

[0012] Figure 2is a flowchart of an object recommendation method provided by an embodiment of the present disclosure;

[0013] Figure 3 is a structural schematic diagram of an object recommendation device provided by an embodiment of the present disclosure;

[0014] Figure 4 It is a structural schematic diagram of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0015] In the following description, specific details such as specific system structures and technologies are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present disclosure. However, it should be clear to those skilled in the art that the present disclosure may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obstructing the description of the present disclosure with unnecessary details.

[0016] It should be noted that the user information (including but not limited to terminal device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0017] A method and device for recommending an object according to an embodiment of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0018] Figure 1 Schematic diagram of an application scenario of an embodiment of the present disclosure. The application scenario may include terminal devices 1, 2 and 3, a server 4 and a network 5.

[0019] Terminal devices 1, 2 and 3 can be hardware or software. When terminal devices 1, 2 and 3 are hardware, they can be various electronic devices with display screens and supporting communication with server 4, including but not limited to smart phones, tablet computers, laptop portable computers and desktop computers, etc.; when terminal devices 1, 2 and 3 are software, they can be installed in the above electronic devices. Terminal devices 1, 2 and 3 can be implemented as multiple software or software modules, or as a single software or software module, and the embodiments of the present disclosure are not limited to this. Furthermore, various applications can be installed on terminal devices 1, 2 and 3, such as data processing applications, instant messaging tools, social platform software, search applications, shopping applications, etc.

[0020] The server 4 may be a server that provides various services, for example, a background server that receives a request sent by a terminal device that establishes a communication connection with the server, and the background server may receive and analyze the request sent by the terminal device, and generate a processing result. The server 4 may be a single server, or a server cluster composed of several servers, or a cloud computing service center, which is not limited in the embodiments of the present disclosure.

[0021] It should be noted that the server 4 can be hardware or software. When the server 4 is hardware, it can be various electronic devices that provide various services for the terminal devices 1, 2, and 3. When the server 4 is software, it can be multiple software or software modules that provide various services for the terminal devices 1, 2, and 3, or it can be a single software or software module that provides various services for the terminal devices 1, 2, and 3, and the embodiments of the present disclosure are not limited to this.

[0022] The network 5 can be a wired network connected by coaxial cable, twisted pair and optical fiber, or it can be a wireless network that can interconnect various communication devices without wiring, such as Bluetooth, Near Field Communication (NFC), infrared, etc., which is not limited in the embodiments of the present disclosure.

[0023] The user can establish a communication connection with the server 4 via the terminal devices 1, 2 and 3 via the network 5 to receive or send information, etc. Specifically, the server 4 can obtain the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended authorized by the user via the terminal devices 1, 2 and 3, and input the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended into the pre-trained object recommendation model for subsequent processing, perform session division on the user historical behavior feature vector, divide the user historical behavior feature vector into at least one user historical session vector, and then process each user historical session vector and the attribute vector corresponding to the object to be recommended through the attention mechanism, and obtain the self-attention vector corresponding to each user historical session vector after processing through the attention mechanism, and then process each user historical session vector and the self-attention vector corresponding to each user historical session vector through the gating mechanism to obtain each user historical session vector. The gating vector corresponding to the quantity can be obtained, and the gating vectors corresponding to each user historical session vector can be fused to obtain a user interest vector. The user interest vector can be used to characterize the user's interest changes over a period of time, and then the recommendation score corresponding to the object to be recommended can be calculated according to the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended. The recommendation score can be used to characterize the user's interest in the object to be recommended, and the recommendation score corresponding to the object to be recommended can be judged and screened based on a preset recommendation score threshold. If the recommendation score corresponding to the object to be recommended is greater than or equal to the preset recommendation score threshold, the object to be recommended can be determined as the target object data that can be displayed to the user, and the target object data can be sent to the target terminal device used by the user, for displaying the target object data on the current graphical user interface of the target terminal device.

[0024] It should be noted that the specific types, quantities and combinations of the terminal devices 1, 2 and 3, the server 4 and the network 5 can be adjusted according to the actual needs of the application scenario, and the embodiments of the present disclosure are not limited to this.

[0025] Figure 2 It is a flowchart of an object recommendation method provided by an embodiment of the present disclosure. Figure 2 The object recommendation method can be Figure 1 The server executes. Figure 2 As shown, the object recommendation method includes:

[0026] Step 201, obtain a user attribute vector, a user historical behavior feature vector, and an attribute vector corresponding to the object to be recommended, and input the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended into a trained object recommendation model.

[0027] Specifically, the user's attribute vector, the user's historical behavior feature vector, and the attribute vector corresponding to the object to be recommended can be obtained through data collection and processing, and the obtained vector can be sent as input data to the trained object recommendation model for recommendation calculation. The user's attribute vector can be the result of converting the user's basic information, which includes but is not limited to age, gender, and / or occupation. The user's historical behavior feature vector can be used to characterize the user's past behavior records, such as purchase records and / or browsing records, which are not limited here. The object to be recommended includes but is not limited to the price, type, brand, and / or duration and name of the new media content of the product, thereby enhancing the personalization of the recommendation, improving the accuracy of the object recommendation model, and improving the user experience.

[0028] Among them, the user attribute vector can be a numerical vector obtained by transforming the user's basic information. The user attribute vector can be used to characterize the static characteristics of the user. The transformation process can be performed by one-hot encoding, embedded vectorization, etc., which is not limited here. The user historical behavior feature vector can be a numerical vector transformed from the user's past behavior records. The transformation can be achieved by behavior frequency statistics or behavior sequence modeling, etc., which is not limited here. The user historical behavior feature vector can be used to characterize the user's dynamic characteristics and preferences. The object to be recommended can be a commodity, new media content, or a regular entity that needs to be recommended. The attribute vector corresponding to the object to be recommended can be a numerical vector transformed from the characteristic attributes of the object to be recommended. The transformation can be achieved by attribute extraction or feature engineering, etc., which is not limited here. The attribute vector corresponding to the object to be recommended can be used to characterize the characteristics and attributes of the object to be recommended.

[0029] The object recommendation model can be a model trained by a machine learning algorithm, and the machine learning algorithm includes but is not limited to collaborative filtering or deep learning. The object recommendation model can learn and extract the potential relationship between the user and the object to be recommended based on the input user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended. The object recommendation model can be used in a variety of recommendation scenarios, including but not limited to e-commerce recommendations, media content recommendations, or text content recommendations.

[0030] For example, on an e-commerce platform, when a user browses or searches for products, the object recommendation model can obtain the user's attribute vector, such as age and gender, the user's historical behavior feature vector, such as the category and price range of products purchased in the past, and can obtain the attribute vector of the product to be recommended, such as a newly listed dress, such as price, material, and size. The obtained vector can be input into the trained object recommendation model, and the object recommendation model can recommend products that may be of interest to the user based on the degree of match between the user and the product.

[0031] Step 202: Perform session division processing on the user's historical behavior feature vector to obtain at least one user's historical session vector.

[0032] Specifically, the user's historical behavior feature vector can be processed by session division. Based on the timestamp information of the user behavior and the preset division time period, the user's historical behavior can be divided into at least one independent session. Each session can include the user's continuous behavior within the preset time period. The user's historical behavior feature vector can be divided into at least one user historical session vector through division. Each historical session vector can be used to characterize the user's behavior characteristics within a preset time period in the past, thereby enhancing the timeliness of the object recommendation model, improving the object recommendation model's ability to capture the user's current interests and needs, and improving the accuracy of the recommendation and user satisfaction. Through session division, the user's behavior patterns and interest preferences in different time periods can be more accurately understood.

[0033] The user history session vector may be a numerical vector corresponding to the independent sessions into which the user's historical behavior is divided within a preset time period. The user history session vector may be obtained by aggregating or encoding all the user's behavior features during the session. The user history session vector may be used to characterize the user's behavior pattern and interest preference within a specific time period. The user history session vector may be obtained in a variety of ways, including but not limited to vector representation based on behavior frequency statistics or vector representation based on behavior sequence modeling.

[0034] For example, in a short video recommendation system, the object recommendation model can divide the user's viewing history into multiple sessions based on the timestamp information of the user watching the video. Each session contains the videos that the user watched continuously during a viewing process. The video features in each session can then be aggregated or encoded to obtain the user's historical session vector.

[0035] Step 203: Process each user's historical conversation vector and the attribute vector corresponding to the object to be recommended based on the attention mechanism to obtain a self-attention vector corresponding to each user's historical conversation vector.

[0036] Specifically, based on the attention mechanism, each user's historical conversation vector and the attribute vector corresponding to the object to be recommended can be processed. For example, the degree of attention of each user's historical conversation vector to the attribute vector of the object to be recommended, that is, the attention weight value, is calculated, and each user's historical conversation vector is weighted summed or weighted transformed based on the attention weight value to obtain the self-attention vector corresponding to each user's historical conversation vector. The self-attention vector corresponding to the user's historical conversation vector can be used to characterize the correlation between the user's historical conversation and the object to be recommended, and highlights the part of the user's historical behavior that is most relevant to the object to be recommended, thereby enhancing the interpretability of the object recommendation model, improving the rationality of the recommendation results, and improving the accuracy of the recommendation and user satisfaction. By introducing the attention mechanism, the association between the user's historical conversation and the object to be recommended can be displayed, so that the recommendation results are more in line with the user's expectations and interests. The attention mechanism can highlight the part of the user's historical behavior that is most relevant to the object to be recommended, thereby improving the accuracy of the recommendation.

[0037] Among them, the self-attention vector corresponding to the user's historical conversation vector can be a numerical vector obtained by processing through the attention mechanism. The self-attention vector corresponding to the user's historical conversation vector can be used to characterize the correlation between the user's historical conversation and the object to be recommended. The self-attention vector corresponding to the user's historical conversation vector can be obtained by calculating the attention weight value of the user's historical conversation vector to the attribute vector of the object to be recommended, and performing weighted summation or weighted transformation on the user's historical conversation vector based on the attention weight value.

[0038] For example, in the product recommendation system on the e-commerce platform, the object recommendation model can divide the user's behavior into multiple sessions based on the user's historical purchasing and browsing behavior, and then for each product to be recommended, the attention weight value between the user's historical session vector and the product attribute vector can be calculated based on the attention mechanism, and the user's historical session vector can be weighted summed based on the attention weight value to obtain the self-attention vector corresponding to each session.

[0039] Step 204 , performing gating mechanism processing on each user historical conversation vector and the self-attention vector corresponding to each user historical conversation vector, to obtain a gating vector corresponding to each user historical conversation vector.

[0040] Specifically, each user historical conversation vector and the self-attention vector corresponding to each user historical conversation vector can be processed by a gating mechanism. The user historical conversation vector and the self-attention vector can be received as input through a gating network, and a gating coefficient can be output. The gating coefficient can be used to control the contribution of the user historical conversation vector in subsequent recommendation calculations, that is, the user historical conversation vector can be weighted to obtain the gating vector corresponding to each user historical conversation vector. The gating mechanism can dynamically adjust the importance of the user historical conversation vector in the recommendation process according to the information of the self-attention vector, thereby enhancing the flexibility of the object recommendation model, improving the adaptability of the recommendation results to the user's current interests and needs, and improving the personalization and accuracy of the recommendation. By introducing the gating mechanism, the object recommendation model can dynamically adjust the weight of the user historical conversation vector in the recommendation calculation according to the correlation between the user's historical conversation and the object to be recommended, so that the recommendation result is more in line with the user's current needs and interests.

[0041] Among them, the gating vector corresponding to the user's historical session vector can be a numerical vector obtained by processing through a gating mechanism. The gating vector corresponding to the user's historical session vector can be used to characterize the contribution of the user's historical session vector in the recommendation calculation. The gating vector can be obtained by jointly processing the user's historical session vector and the self-attention vector through a gating network. The gating vector can contain information about the importance of the user's historical session vector in the recommendation process and can serve as an important input for subsequent recommendation calculations.

[0042] For example, in a news recommendation system, the object recommendation model can divide the user's behavior into multiple sessions based on the user's historical reading behavior. For each news to be recommended, the object recommendation model can calculate the attention weight value between the user's historical session vector and the news attribute vector based on the attention mechanism, and perform weighted summation of the user's historical session vectors based on the attention weight value to obtain the self-attention vector corresponding to each session. The object recommendation model can process each session vector and the corresponding self-attention vector through a gating mechanism to obtain a gating vector corresponding to each session. The gating vector can be used to characterize the user's preference and interest in different types of news in different sessions, and dynamically adjust the weight of each session vector in the recommendation calculation. For example, if the user frequently reads science and technology news in a session, the object recommendation model can give the session a higher weight value in the recommendation calculation based on the gating vector of the session, so that more science and technology news can be recommended to the user, thereby improving the personalization and accuracy of the recommendation results.

[0043] Step 205 , fusing the gating vectors corresponding to the historical conversation vectors of each user to obtain a user interest vector.

[0044] Specifically, the gating vectors corresponding to the historical session vectors of each user can be fused, which can be achieved by weighted averaging, splicing, or fusion of the gating vectors through a neural network, etc., which are not limited here. A user interest vector can be obtained, and the user interest vector can be used to characterize the interest preferences of the user in different historical sessions, thereby enhancing the ability of the object recommendation model to capture user interests, improving the comprehensiveness and accuracy of the recommendation results, and enhancing the user experience. By fusing the gating vectors corresponding to the historical session vectors of each user, the interest preferences of users in different time periods and different situations can be comprehensively considered, so as to recommend content that is more in line with their overall interests to users.

[0045] Among them, the user interest vector can be a numerical vector obtained through fusion processing, and the user interest vector can be used to characterize the user's interest characteristics. The user interest vector can be obtained by weighted averaging, splicing, or neural network fusion of the gating vectors corresponding to each user's historical session vector. The user interest vector can include the user's interest preference information in different sessions and can be used as an input for subsequent recommendation calculations to generate personalized recommendation results.

[0046] For example, in a music recommendation system, the object recommendation model can divide the user's behavior into multiple sessions based on the user's historical listening behavior, and then fuse the gating vectors corresponding to each session to obtain the user's interest vector. The user's interest vector may include the user's preferences for music in different time periods and different situations, such as preferences for different music types such as pop, classical, and / or rock. If the user listens to pop music multiple times in different sessions, the object recommendation model can recommend more pop songs to the user, thereby improving the user experience.

[0047] Step 206: Determine the recommendation score corresponding to the object to be recommended based on the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended.

[0048] Specifically, the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended can be jointly processed, and the recommendation score corresponding to the object to be recommended can be obtained based on the user's basic attributes, historical interests, and the characteristics of the object to be recommended. The recommendation score corresponding to the object to be recommended can be used to characterize the degree of match between the object to be recommended and the user's interests and needs, thereby enhancing the judgment ability of the object recommendation model, improving the pertinence and accuracy of the recommendation results, and improving the user experience and satisfaction. Through the joint processing of user attributes, user interests and the attributes of the object to be recommended, the object recommendation model can more comprehensively evaluate the degree of match between the object to be recommended and the user, thereby recommending content that is more in line with the user's interests and needs.

[0049] Among them, the recommendation score corresponding to the object to be recommended can be a numerical value used to characterize the degree of match between the object to be recommended and the user's interests and needs. The recommendation score corresponding to the object to be recommended can be calculated by inputting the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended into the object recommendation model. The higher the recommendation score, the higher the degree of match between the object to be recommended and the user, and the more interested the user is in the object to be recommended.

[0050] For example, in a movie recommendation system, the object recommendation model can be based on the user's age, gender and other attribute information, as well as the user's historical viewing behavior and interest preferences combined with the movie's type, director, and actors and other attribute information, and the above information is input into the object recommendation model to calculate the recommendation score of each movie for the user.

[0051] Step 207, based on the preset recommendation score threshold and the recommendation score corresponding to the object to be recommended, determine the target object data to be displayed to the user, and send the target object data to the target terminal device so that the target object data is displayed on the current graphical user interface of the target terminal device.

[0052] Specifically, by comparing the recommendation scores corresponding to each object to be recommended with a preset recommendation score threshold, the objects to be recommended whose recommendation scores are greater than or equal to the preset recommendation score threshold can be screened out as target objects, and then the target object data can be sent to the target terminal device. The target terminal device can display the target object data on the user's current graphical user interface for the user to view and select, wherein the target terminal device includes but is not limited to the user's mobile phone or computer, and the current graphical user interface can be the interface that the user is currently using, thereby enhancing the practicality and user experience of the object recommendation model, improving the effectiveness and conversion rate of the recommendation results, and improving the interactivity and information display capabilities of the target terminal device. By using the preset recommendation score threshold and screening the target object data, it can be ensured that content that meets the user's interests and needs is displayed to the user, thereby improving user satisfaction and engagement.

[0053] Among them, the preset recommendation score threshold can be a pre-set numerical standard for screening objects to be recommended. The preset recommendation score threshold can be set according to the recommendation strategy and the needs of the user. The preset recommendation score threshold can also be used to determine the objects to be recommended that are worth displaying to the user. The target object data can refer to the relevant information of the objects to be recommended that are screened and have a recommendation score greater than or equal to the preset recommendation score threshold. The relevant information includes but is not limited to the title, picture, and / or introduction, etc. The target object data can be used to display to the user on the graphical user interface of the target terminal device. The target terminal device can be a device that the user is currently using, including but not limited to a mobile phone or a computer, etc. The target terminal device can be used to receive and display the target object data. The current graphical user interface can be the interface that the user is viewing on the target terminal device. The current graphical user interface provides a window for interacting with the user, which can be used to display recommended content.

[0054] For example, in the product recommendation system of an e-commerce platform, the object recommendation model can calculate the recommendation score of each product based on the user's shopping history and interest preferences, and can filter out products with recommendation scores greater than or equal to the threshold as target objects based on a preset recommendation score threshold. By obtaining information such as pictures, titles, and / or prices of the target products and sending them to the user's mobile phone or computer and other target terminal devices, the target terminal device displays the product information on its current graphical user interface, and the user can see the recommended products and click to enter the details page to purchase.

[0055] According to the technical solution provided by the embodiment of the present disclosure, by obtaining a user attribute vector, a user historical behavior feature vector, and an attribute vector corresponding to an object to be recommended that are authorized by the user, and inputting the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended into a pre-trained object recommendation model for subsequent processing, the user historical behavior feature vector is divided into sessions, and the user historical behavior feature vector is divided into at least one user historical session vector, and then each user historical session vector and the attribute vector corresponding to the object to be recommended can be processed through an attention mechanism, and a self-attention vector corresponding to each user historical session vector can be obtained after processing through the attention mechanism, and then each user historical session vector and the self-attention vector corresponding to each user historical session vector can be processed through a gating mechanism to obtain a gating vector corresponding to each user historical session vector, and the gating vectors corresponding to each user historical session vector can be fused to obtain a user interest vector, which can be used to characterize the user's interest in a period of time. The recommendation score corresponding to the object to be recommended can be calculated according to the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended. The recommendation score can be used to characterize the user's interest in the object to be recommended, and the recommendation score corresponding to the object to be recommended can be judged and screened based on a preset recommendation score threshold. If the recommendation score corresponding to the object to be recommended is greater than or equal to the preset recommendation score threshold, the object to be recommended can be determined as the target object data that can be displayed to the user, and the target object data can be sent to the target terminal device used by the user, so as to display the target object data on the current graphical user interface of the target terminal device. In this way, the personalization of the recommendation is improved by considering the user attribute vector and the user's historical behavior feature vector. Through the processing based on the attention mechanism, the object recommendation model's ability to capture the user's dynamic interests is improved. The gating mechanism is introduced to process the user's historical session vector and the corresponding self-attention vector, which reduces noise interference and enhances the scalability and flexibility of the object recommendation model.

[0056] In some embodiments, obtaining a user attribute vector, a user historical behavior feature vector, and an attribute vector corresponding to the object to be recommended, and before inputting the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended into a trained object recommendation model, also includes: obtaining user attribute data, user historical behavior data, and attribute data corresponding to the object to be recommended; and embedding the user attribute data, the user historical behavior data, and the attribute data corresponding to the object to be recommended, respectively, to obtain the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended.

[0057] Specifically, user attribute data, user historical behavior data, and attribute data corresponding to the object to be recommended can be obtained from multiple data sources. The user attribute data includes but is not limited to basic information such as the user's age, gender, and / or occupation. The user historical behavior data includes but is not limited to the user's past purchase records, browsing records, and / or search records. The attribute data corresponding to the object to be recommended includes but is not limited to the category, brand, price, and / or evaluation of the product. The obtained data can then be preprocessed, including but not limited to data cleaning and / or format conversion. The processed data can be embedded and converted into vector form. The embedding processing includes but is not limited to word embedding or graph embedding. Discrete input data can be converted into a continuous vector representation. The user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended can be obtained through the embedding processing, and the aforementioned vectors are input into the trained object recommendation model.

[0058] Among them, user attribute data can be information used to describe the basic characteristics of the user, including but not limited to age, gender, occupation, and / or region, etc., and user attribute data can be used to characterize the static characteristics of the user. User historical behavior data can be a record of the user's behavior in the past period of time, including but not limited to purchase records, browsing records, and / or search records, etc., and user historical behavior data can be used to characterize the user's dynamic interests and preferences. The attribute data corresponding to the object to be recommended can be information used to describe the characteristics of the object to be recommended, including but not limited to the category, brand, price, and / or evaluation of the product, etc., and the attribute data corresponding to the object to be recommended can be used to characterize the attributes and characteristics of the object to be recommended.

[0059] The current graphical user interface (GUI) can be an interface displayed on the device that the user is currently using. The GUI can provide windows and interface elements for interacting with the user, including but not limited to buttons, text boxes, and / or pictures. The current graphical user interface can be used to display recommendation results to the user and allow the user to perform subsequent operations, including but not limited to clicking and / or purchasing.

[0060] For example, in the product recommendation system of an e-commerce platform, the object recommendation model can obtain attribute data such as the user's age and / or gender through the user registration information after the user's authorization, obtain the user's historical behavior data through the user's shopping records, and obtain attribute data such as the product category and / or price through product information, and then embed the obtained data to obtain a user attribute vector, a user historical behavior feature vector, and a product attribute vector.

[0061] According to the technical solution provided by the embodiments of the present disclosure, user attribute data, user historical behavior data, and attribute data corresponding to the object to be recommended are obtained through multiple data sources, and the obtained data can be preprocessed. The processed data can be embedded and converted into vector form respectively, and discrete input data can be converted into continuous vector representation. User attribute vectors, user historical behavior feature vectors, and attribute vectors corresponding to the object to be recommended can be obtained through embedding processing, thereby enhancing the data processing capability of the object recommendation model, improving the quality and availability of data, and improving the accuracy and efficiency of the object recommendation model. By embedding the user attribute data, user historical behavior data, and attribute data corresponding to the object to be recommended, the original data can be converted into a vector form that is more suitable for recommendation calculation, thereby improving the calculation efficiency and accuracy of the object recommendation model.

[0062] In some embodiments, the user historical behavior feature vectors are processed for session division to obtain at least one user historical session vector, including: concatenating the user historical behavior feature vectors to obtain at least one user historical behavior vector sequence; and dividing each user historical behavior vector sequence based on a preset time node to obtain at least one user historical session vector.

[0063] Specifically, the user historical behavior feature vectors may be concatenated in chronological order to obtain one or more user historical behavior vector sequences, each of which may be used to characterize the user's behavior characteristics at a certain point in time or during a period of time. The object recommendation model may be based on preset time nodes, including but not limited to midnight of each day or the end of each week, to divide the user historical behavior vector sequence, and the vectors before and after the time nodes may be respectively classified into different sessions, thereby obtaining at least one user historical session vector, each of which may contain the user's behavior feature vector during a certain session, and the user historical session vector may be used to characterize the user's overall behavior pattern in the session.

[0064] The user historical behavior vector sequence may be a set of user historical behavior feature vectors arranged in chronological order. The user historical behavior vector sequence may be used to characterize the user's behavior records generated within a period of time, and may also be used to characterize the user's dynamic interests and preferences. The preset time node may be a time point or time period used to divide the user historical behavior vector sequence. The preset time node may be set according to the recommendation strategy and the user's behavior characteristics, including but not limited to midnight every day or the end of each week. By setting the preset time node, the object recommendation model may divide the user's behavior into different session units.

[0065] For example, in the content recommendation system of a video platform, the object recommendation model can obtain the historical behavior feature vectors of users watching videos, and splice them according to the chronological order to obtain the user's historical behavior vector sequence. Then, midnight of each day can be set as the preset time node, and the user's historical behavior vector sequence can be divided into multiple user historical session vectors. Each session vector can contain the behavior feature vectors of all videos watched by the user in one day.

[0066] According to the technical solution provided by the embodiments of the present disclosure, by splicing the user historical behavior feature vectors according to the time sequence, one or more user historical behavior vector sequences are obtained. The object recommendation model can divide the user historical behavior vector sequence based on the preset time node, and can classify the vectors before and after the time node into different sessions respectively, so as to obtain at least one user historical session vector, thereby enhancing the object recommendation model's ability to parse the user's historical behavior, improving the accuracy of capturing user behavior patterns, and improving the personalization and accuracy of recommendation results. By dividing the user's historical behavior feature vectors into sessions, the object recommendation model can organize the user's behavior according to the time sequence and session units, so as to better understand the user's behavior preferences in different time periods and situations.

[0067] In some embodiments, each user historical conversation vector and the attribute vector corresponding to the object to be recommended are processed based on the attention mechanism to obtain a self-attention vector corresponding to each user historical conversation vector, including: performing weight calculation processing on each user historical conversation vector and the attribute vector corresponding to the object to be recommended to obtain a weight value corresponding to each user historical conversation vector; performing feature fusion processing on the weight value corresponding to each user historical conversation vector and each user historical conversation vector to obtain a self-attention vector corresponding to each user historical conversation vector.

[0068] Specifically, each historical conversation vector and the attribute vector corresponding to the object to be recommended can be processed through the attention mechanism, the correlation between each user historical conversation vector and the attribute vector corresponding to the object to be recommended can be calculated, and a weight value can be assigned to each user historical conversation vector based on the correlation, that is, the weight value corresponding to each user historical conversation vector, and the weight value can be used to characterize the importance or relevance of the historical conversation vector to the current object to be recommended, and then the weight value can be feature fused with the corresponding user historical conversation vector, that is, the weight value can be used as a weighting factor and multiplied with each feature dimension of the user historical conversation vector to obtain the self-attention vector corresponding to each user historical conversation vector.

[0069] Among them, the weight value corresponding to the user's historical conversation vector can be a correlation value calculated by the attention mechanism. The weight value corresponding to the user's historical conversation vector can be used to characterize the correlation between the user's historical conversation vector and the attribute vector corresponding to the object to be recommended. The weight value can be obtained by calculation methods such as dot product or cosine similarity. The weight value can be used to measure the similarity or correlation between the user's historical conversation vector and the object to be recommended. The larger the weight value, the higher the correlation between the conversation vector and the object to be recommended, and vice versa.

[0070] For example, in the product recommendation system of an e-commerce platform, the object recommendation model can obtain the user's historical shopping session vector and the attribute vector corresponding to the product to be recommended, and can calculate the weight value between each historical shopping session vector and the attribute vector of the product to be recommended through the attention mechanism. The weight value can be used to characterize the user's attention and interest in the current product to be recommended in different shopping sessions. Then, the object recommendation model can perform feature fusion processing on the weight value and the corresponding historical shopping session vector to obtain the self-attention vector corresponding to each session vector.

[0071] According to the technical solution provided by the embodiment of the present disclosure, each historical session vector and the attribute vector corresponding to the object to be recommended are processed through the attention mechanism, the correlation between each user historical session vector and the attribute vector corresponding to the object to be recommended can be calculated, and a weight value can be assigned to each user historical session vector based on the correlation, that is, the weight value corresponding to each user historical session vector, and then the weight value can be subjected to feature fusion processing with the corresponding user historical session vector, that is, the weight value can be used as a weighting factor and multiplied with each feature dimension of the user historical session vector to obtain the self-attention vector corresponding to each user historical session vector, thereby enhancing the sensitivity and distinguishing ability of the object recommendation model to the user's historical session, improving the accuracy of capturing user interests and behavior patterns, and improving the personalization and relevance of the recommendation results. By introducing the attention mechanism and calculating the weight value, the object recommendation model can more accurately evaluate the importance of each user historical session vector to the current object to be recommended, thereby ensuring that the recommendation result is more in line with the user's actual needs and interests, helping to reduce the interference of noise information, and improving the robustness and accuracy of the object recommendation model.

[0072] In some embodiments, each user historical session vector and the self-attention vector corresponding to each user historical session vector are processed by a gating mechanism to obtain a gating vector corresponding to each user historical session vector, including: performing nonlinear activation processing on each user historical session vector and the self-attention vector corresponding to each user historical session vector to obtain a retention vector corresponding to each user historical session vector; processing the retention vector corresponding to each user historical session vector by resetting a gate to obtain a gating vector corresponding to each user historical session vector.

[0073] Specifically, each user historical session vector and the self-attention vector corresponding to each user historical session vector can be nonlinearly activated through the update gate of the gated network. The nonlinear activation processing can be performed through activation functions such as ReLU or sigmoid, which are not limited here, so as to obtain a retained vector corresponding to each user historical session vector. Then, the object recommendation model can process the retained vector through a reset gate in the gated network. The reset gate can be a module composed of a fully connected layer and an activation function. The reset gate can dynamically adjust the degree of retention of output information according to the input information. Through the reset gate processing, the object recommendation model can obtain the gating vector corresponding to each user historical session vector.

[0074] The retained vector corresponding to the user's historical conversation vector may be a vector obtained through nonlinear activation processing, and the retained vector corresponding to the user's historical conversation vector may include historical conversation vector information and self-attention vector correlation information.

[0075] The reset gate can be a gating structure for controlling the flow of information. The reset gate can be composed of a fully connected layer and an activation function. The activation function includes but is not limited to a sigmoid function or a ReLU function. The reset gate can dynamically adjust the degree of retention of the output information according to the retention vector. The reset gate can obtain a weight value between 0 and 1 through calculation, which is the gating coefficient. The gating coefficient can be used to determine the part of the retention vector that can be discarded. By adjusting the gating coefficient, the reset gate can control the flow and retention of information, thereby achieving effective screening and filtering of input information.

[0076] For example, in a news recommendation system, the object recommendation model can obtain the user's historical reading session vector and the attribute vector of the current news to be recommended, calculate the self-attention vector between each historical reading session vector and the attribute vector of the news to be recommended through the attention mechanism of the update gate, and perform nonlinear activation processing on the historical reading session vectors and their corresponding self-attention vectors to obtain a retained vector, which can then be processed by the reset gate to obtain a gated vector.

[0077] According to the technical solution provided by the embodiment of the present disclosure, each user historical session vector and the self-attention vector corresponding to each user historical session vector are non-linearly activated through the update gate of the gated network, which is not limited here, so as to obtain the retention vector corresponding to each user historical session vector, and then the object recommendation model can process the retention vector through the reset gate in the gated network, and the reset gate can dynamically adjust the retention degree of the output information according to the input information. The object recommendation model can obtain the gating vector corresponding to each user historical session vector, thereby enhancing the object recommendation model's screening and filtering capabilities for user historical session information, improving the accuracy and robustness of information representation, and enhancing the personalization and relevance of recommendation results. By introducing the gating mechanism, the object recommendation model can control the flow and retention of information in the user historical session vector, thereby ensuring the practicality of the recommendation results.

[0078] In some embodiments, before obtaining a user attribute vector, a user historical behavior feature vector, and an attribute vector corresponding to an object to be recommended, and inputting the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended into a trained object recommendation model, the method further includes: obtaining user attribute training data, user historical behavior training data, attribute data corresponding to the training object, and a real label corresponding to the training object; embedding the user attribute training data, the user historical behavior training data, and the attribute data corresponding to the training object to obtain a user attribute training vector, a user historical behavior feature training vector, and an attribute vector corresponding to the training object; performing session division processing on the user historical behavior feature training vector through the object recommendation model to obtain at least one user historical session training vector; and performing a segmentation processing on the user historical behavior feature training vector based on an attention machine. The method processes each user history session training vector and the attribute vector corresponding to the training object to obtain the self-attention vector corresponding to each user history session training vector; performs gating mechanism processing on each user history session training vector and the self-attention vector corresponding to each user history session training vector to obtain the gating vector corresponding to each user history session training vector; fuses the gating vectors corresponding to each user history session training vector to obtain the user interest training vector; determines the recommendation score corresponding to the training object according to the user attribute training vector, the user interest training vector, and the attribute vector corresponding to the training object; determines the loss of the object recommendation model according to the recommendation score corresponding to the training object and the true label corresponding to the training object; updates the parameters in the object recommendation model according to the loss corresponding to the object recommendation model through a loop iteration.

[0079] Specifically, user attribute training data, user historical behavior training data, attribute data corresponding to the training object, and the real label corresponding to each training object can be obtained, and then the user attribute training data, user historical behavior training data, and attribute data corresponding to the training object can be embedded respectively to obtain user attribute training vectors, user historical behavior feature training vectors, and attribute vectors corresponding to the training object. The user historical behavior feature training vectors can be divided into sessions through the object recommendation model to obtain at least one user historical session training vector. Based on the attention mechanism, each user historical session training vector and the attribute vector corresponding to the training object are processed to obtain each user historical session training vector. The self-attention vector corresponding to the historical conversation training vector can be obtained, and the gating mechanism can be performed on the user historical conversation training vector and the corresponding self-attention vector to obtain the gating vector corresponding to each user historical conversation training vector. The gating vector can be fused to obtain the user interest training vector. The object recommendation model can determine the recommendation score corresponding to the training object according to the user attribute training vector, the user interest training vector, and the attribute vector corresponding to the training object. The recommendation score corresponding to the training object can be compared with the true label of the training object to calculate the loss of the object recommendation model. The parameters in the object recommendation model can be updated according to the loss value corresponding to the object recommendation model through a loop iteration.

[0080] The user attribute training data may be user attribute information used to train the object recommendation model, including but not limited to the user's age, gender, and / or occupation.

[0081] The user historical behavior training data may be a record of the user's behavior over a period of time in the past, including but not limited to purchase records and / or browsing records, etc. The user historical behavior training data may be used to characterize the user's interests and behavior patterns.

[0082] The training object may be an object to be recommended for training the object recommendation model, and the training object includes but is not limited to commodities, news, or videos, etc. The attribute data contained in the training object includes but is not limited to price, category, and / or title, etc.

[0083] The attribute data corresponding to the training object may be attribute information of the training object, including but not limited to the category, price, and / or brand of the product. The attribute data corresponding to the training object may help the object recommendation model to more accurately understand the relationship between the object and the user.

[0084] The true label corresponding to the training object can be used to characterize whether the training object is liked or selected by the user in actual applications. The true label can be the basis for evaluating the model performance during the object recommendation model training process. The loss value corresponding to the object recommendation model can be calculated by comparing with the recommendation score.

[0085] The loss of the object recommendation model can be the difference between the recommendation score predicted by the object recommendation model during the training process and the true label. The loss value can be used to characterize the performance of the object recommendation model and can be the basis for updating the parameters of the object recommendation model. By reducing the loss value, the performance of the object recommendation model can be optimized and the accuracy and stability of the recommendation can be improved.

[0086] According to the technical solution provided by the embodiment of the present disclosure, by obtaining user attribute training data, user historical behavior training data, attribute data corresponding to the training object, and the real label corresponding to each training object, the user attribute training data, the user historical behavior training data, and the attribute data corresponding to the training object can be embedded respectively to obtain a user attribute training vector, a user historical behavior feature training vector, and an attribute vector corresponding to the training object. The user historical behavior feature training vector can be processed by session division through the object recommendation model to obtain at least one user historical session training vector. Based on the attention mechanism, each user historical session training vector and the attribute vector corresponding to the training object are processed to obtain a self-attention vector corresponding to each user historical session training vector. Then, the user historical session training vector and the corresponding self-attention vector can be processed by a gating mechanism to obtain a gating vector corresponding to each user historical session training vector. The gating vector can be integrated The object recommendation model can determine the recommendation score corresponding to the training object according to the user attribute training vector, the user interest training vector, and the attribute vector corresponding to the training object. The recommendation score corresponding to the training object can be compared with the real label of the training object to calculate the loss of the object recommendation model. The parameters in the object recommendation model can be updated according to the loss value corresponding to the object recommendation model in a loop iteration manner, thereby enhancing the training efficiency and accuracy of the object recommendation model, improving the ability of the object recommendation model to capture user interests and behaviors, and improving the personalization and relevance of the recommendation results. By introducing user attribute training data, user historical behavior training data, and attribute data and real labels corresponding to the training objects, the object recommendation model can learn the relationship between users and objects more comprehensively, thereby generating more accurate recommendation results. The model parameters are continuously updated in a loop iteration manner, thereby improving the accuracy and stability of the recommendation.

[0087] In some embodiments, a recommendation score corresponding to an object to be recommended is determined based on a user attribute vector, a user interest vector, and an attribute vector corresponding to the object to be recommended, including: concatenating the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended to obtain a concatenated vector corresponding to the object to be recommended; flattening the concatenated vector corresponding to the object to be recommended to obtain a low-dimensional vector corresponding to the object to be recommended; mapping the low-dimensional vector corresponding to the object to be recommended through a multi-layer perceptron to obtain a spatial mapping vector corresponding to the object to be recommended; and calculating the recommendation score of the spatial mapping vector corresponding to the object to be recommended to obtain a recommendation score corresponding to the object to be recommended.

[0088] Specifically, the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended can be concatenated to obtain a concatenated vector corresponding to the object to be recommended. The concatenated vector can be flattened to convert the concatenated vector into a low-dimensional vector. The object recommendation model can map the low-dimensional vector through a multi-layer perceptron to obtain a spatial mapping vector corresponding to the object to be recommended. The spatial mapping vector can be a representation of the object to be recommended in a specific space. The spatial mapping vector can be used to characterize the potential relationship between the object to be recommended and the user. Then, the recommendation score calculation can be performed on the spatial mapping vector to obtain a recommendation score corresponding to the object to be recommended.

[0089] The concatenated vector corresponding to the object to be recommended may be a vector obtained by concatenating the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended.

[0090] The low-dimensional vector corresponding to the object to be recommended may be a vector obtained by flattening the concatenated vector corresponding to the object to be recommended. The low-dimensional vector retains the original information and reduces the data dimension, which is convenient for subsequent processing.

[0091] A multilayer perceptron (MLP) can be a feedforward neural network. MLP can be composed of multiple layers, each layer contains multiple neurons, and each neuron can receive input from the neurons in the previous layer and output it to the next layer after nonlinear transformation through an activation function. Multilayer perceptrons can be used to learn complex relationships between input data and map them to output space. Multilayer perceptrons can also be used to map low-dimensional vectors to spatial mapping vectors in a specific space.

[0092] The space mapping vector corresponding to the object to be recommended may be a vector obtained by mapping the low-dimensional vector through a multi-layer perceptron. The space mapping vector may be used to characterize the potential relationship between the object to be recommended and the user in a specific space.

[0093] For example, in the product recommendation system of an e-commerce platform, the object recommendation model can obtain user attribute vectors, such as the user's age, gender, and / or occupation, the user's interest vector, and the attribute vector corresponding to the product to be recommended, such as the product category, price, and / or brand, etc., concatenate the obtained vectors to obtain a concatenated vector, and then flatten the concatenated vector to obtain a low-dimensional vector. The low-dimensional vector can then be mapped through a multi-layer perceptron to obtain a spatial mapping vector, and the recommendation score of the product to be recommended is calculated based on the spatial mapping vector, and the products are sorted and recommended based on the recommendation score.

[0094] According to the technical solution provided by the embodiment of the present disclosure, a user attribute vector, a user interest vector, and an attribute vector corresponding to the object to be recommended are concatenated to obtain a concatenated vector corresponding to the object to be recommended. The concatenated vector can be flattened to convert the concatenated vector into a low-dimensional vector. The object recommendation model can map the low-dimensional vector through a multi-layer perceptron to obtain a spatial mapping vector corresponding to the object to be recommended. The spatial mapping vector can then be processed for recommendation score calculation to obtain a recommendation score corresponding to the object to be recommended, thereby enhancing the processing capability and flexibility of the object recommendation model, improving the accuracy and relevance of the recommendation results, and improving the user experience and satisfaction. Through multiple steps such as concatenation, flattening, multi-layer perceptron mapping, and recommendation score calculation, the object recommendation model can more comprehensively consider multiple factors such as user attributes, user interests, and attributes of the object to be recommended, thereby obtaining more accurate recommendation results.

[0095] All the above optional technical solutions can be arbitrarily combined to form optional embodiments of the present disclosure, and will not be described in detail here.

[0096] The following are embodiments of the device disclosed herein, which can be used to execute the method embodiments disclosed herein. For details not disclosed in the device embodiments disclosed herein, please refer to the method embodiments disclosed herein.

[0097] Figure 3 is a schematic diagram of an object recommendation device provided by an embodiment of the present disclosure. Figure 3 As shown, the object recommendation device includes:

[0098] An acquisition module 301 is used to acquire a user attribute vector, a user historical behavior feature vector, and an attribute vector corresponding to the object to be recommended, and input the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended into a trained object recommendation model;

[0099] The first processing module 302 is used to perform session division processing on the user's historical behavior feature vector to obtain at least one user's historical session vector;

[0100] The second processing module 303 is used to process each user's historical conversation vector and the attribute vector corresponding to the object to be recommended based on the attention mechanism to obtain a self-attention vector corresponding to each user's historical conversation vector;

[0101] The third processing module 304 is used to perform a gating mechanism processing on each user historical conversation vector and the self-attention vector corresponding to each user historical conversation vector to obtain a gating vector corresponding to each user historical conversation vector;

[0102] The fourth processing module 305 is used to fuse the gating vectors corresponding to the historical conversation vectors of each user to obtain a user interest vector;

[0103] A determination module 306, configured to determine a recommendation score corresponding to the object to be recommended based on the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended;

[0104] The fifth processing module 307 is used to determine the target object data to be displayed to the user based on the preset recommendation score threshold and the recommendation score corresponding to the object to be recommended, and send the target object data to the target terminal device so that the target object data is displayed on the current graphical user interface of the target terminal device.

[0105] According to the technical solution provided by the embodiment of the present disclosure, by obtaining a user attribute vector, a user historical behavior feature vector, and an attribute vector corresponding to an object to be recommended that are authorized by the user, and inputting the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended into a pre-trained object recommendation model for subsequent processing, the user historical behavior feature vector is divided into sessions, and the user historical behavior feature vector is divided into at least one user historical session vector, and then each user historical session vector and the attribute vector corresponding to the object to be recommended can be processed through an attention mechanism, and a self-attention vector corresponding to each user historical session vector can be obtained after processing through the attention mechanism, and then each user historical session vector and the self-attention vector corresponding to each user historical session vector can be processed through a gating mechanism to obtain a gating vector corresponding to each user historical session vector, and the gating vectors corresponding to each user historical session vector can be fused to obtain a user interest vector, which can be used to characterize the user's interest in a period of time. The recommendation score corresponding to the object to be recommended can be calculated according to the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended. The recommendation score can be used to characterize the user's interest in the object to be recommended, and the recommendation score corresponding to the object to be recommended can be judged and screened based on a preset recommendation score threshold. If the recommendation score corresponding to the object to be recommended is greater than or equal to the preset recommendation score threshold, the object to be recommended can be determined as the target object data that can be displayed to the user, and the target object data can be sent to the target terminal device used by the user, so as to display the target object data on the current graphical user interface of the target terminal device. In this way, the personalization of the recommendation is improved by considering the user attribute vector and the user's historical behavior feature vector. Through the processing based on the attention mechanism, the object recommendation model's ability to capture the user's dynamic interests is improved. The gating mechanism is introduced to process the user's historical session vector and the corresponding self-attention vector, which reduces noise interference and enhances the scalability and flexibility of the object recommendation model.

[0106] In some embodiments, the above-mentioned object recommendation device is also used to obtain user attribute data, user historical behavior data, and attribute data corresponding to the object to be recommended; the user attribute data, user historical behavior data, and attribute data corresponding to the object to be recommended are respectively embedded to obtain a user attribute vector, a user historical behavior feature vector, and an attribute vector corresponding to the object to be recommended.

[0107] In some embodiments, the first processing module 302 is specifically used to concatenate user historical behavior feature vectors to obtain at least one user historical behavior vector sequence; and divide each user historical behavior vector sequence based on a preset time node to obtain at least one user historical session vector.

[0108] In some embodiments, the second processing module 303 is specifically used to perform weight calculation processing on each user historical conversation vector and the attribute vector corresponding to the object to be recommended, so as to obtain the weight value corresponding to each user historical conversation vector; perform feature fusion processing on the weight value corresponding to each user historical conversation vector and each user historical conversation vector, so as to obtain the self-attention vector corresponding to each user historical conversation vector.

[0109] In some embodiments, the third processing module 304 is specifically used to perform nonlinear activation processing on each user historical session vector and the self-attention vector corresponding to each user historical session vector to obtain a retention vector corresponding to each user historical session vector; and process the retention vector corresponding to each user historical session vector by resetting the gate to obtain a gating vector corresponding to each user historical session vector.

[0110] In some embodiments, the object recommendation device is further used to obtain user attribute training data, user historical behavior training data, attribute data corresponding to the training object, and the real label corresponding to the training object; embed the user attribute training data, the user historical behavior training data, and the attribute data corresponding to the training object to obtain a user attribute training vector, a user historical behavior feature training vector, and an attribute vector corresponding to the training object; perform session division processing on the user historical behavior feature training vector through the object recommendation model to obtain at least one user historical session training vector; process each user historical session training vector and the attribute vector corresponding to the training object based on the attention mechanism to obtain each user historical session training vector. The method comprises the following steps: obtaining a self-attention vector corresponding to a conversation training vector; performing a gating mechanism processing on each user history conversation training vector and the self-attention vector corresponding to each user history conversation training vector to obtain a gating vector corresponding to each user history conversation training vector; fusing the gating vectors corresponding to each user history conversation training vector to obtain a user interest training vector; determining a recommendation score corresponding to the training object according to the user attribute training vector, the user interest training vector, and the attribute vector corresponding to the training object; determining the loss of the object recommendation model according to the recommendation score corresponding to the training object and the true label corresponding to the training object; and updating the parameters in the object recommendation model according to the loss corresponding to the object recommendation model through a loop iteration.

[0111] In some embodiments, the determination module 306 is specifically used to perform splicing processing on the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended to obtain a splicing vector corresponding to the object to be recommended; flatten the splicing vector corresponding to the object to be recommended to obtain a low-dimensional vector corresponding to the object to be recommended; map the low-dimensional vector corresponding to the object to be recommended through a multi-layer perceptron to obtain a spatial mapping vector corresponding to the object to be recommended; and calculate the recommendation score of the spatial mapping vector corresponding to the object to be recommended to obtain a recommendation score corresponding to the object to be recommended.

[0112] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present disclosure.

[0113] Figure 4 Schematic diagram of an electronic device 4 provided in an embodiment of the present disclosure. Figure 4 As shown, the electronic device 4 of this embodiment includes: a processor 401, a memory 402, and a computer program 403 stored in the memory 402 and executable on the processor 401. When the processor 401 executes the computer program 403, the steps in the above-mentioned method embodiments are implemented. Alternatively, when the processor 401 executes the computer program 403, the functions of the modules / units in the above-mentioned device embodiments are implemented.

[0114] The electronic device 4 may be a desktop computer, a notebook, a PDA, a cloud server, or other electronic device. The electronic device 4 may include, but is not limited to, a processor 401 and a memory 402. Those skilled in the art will appreciate that Figure 4 The electronic device 4 is merely an example and does not limit the electronic device 4 , and may include more or less components than those shown in the figure, or different components.

[0115] The processor 401 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0116] The memory 402 may be an internal storage unit of the electronic device 4, for example, a hard disk or memory of the electronic device 4. The memory 402 may also be an external storage device of the electronic device 4, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 4. The memory 402 may also include both an internal storage unit and an external storage device of the electronic device 4. The memory 402 is used to store computer programs and other programs and data required by the electronic device.

[0117] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units.

[0118] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium (e.g., a computer-readable storage medium). Based on this understanding, the present disclosure implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. The computer program may include computer program code, and the computer program code may be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable storage medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0119] The above embodiments are only used to illustrate the technical solutions of the present disclosure, rather than to limit them. Although the present disclosure has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should all be included in the protection scope of the present disclosure.

Claims

1. An object recommendation method, characterized in that: include: Obtaining a user attribute vector, a user historical behavior feature vector, and an attribute vector corresponding to the object to be recommended, and inputting the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended into a trained object recommendation model; Performing session division processing on the user historical behavior feature vector to obtain at least one user historical session vector; Based on the attention mechanism, each of the user's historical conversation vectors and the attribute vector corresponding to the object to be recommended are processed to obtain a self-attention vector corresponding to each of the user's historical conversation vectors; Performing gating mechanism processing on each of the user history conversation vectors and the self-attention vectors corresponding to each of the user history conversation vectors to obtain a gating vector corresponding to each of the user history conversation vectors; Fusing the gating vectors corresponding to the user history session vectors to obtain a user interest vector; Determining a recommendation score corresponding to the object to be recommended according to the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended; Based on a preset recommendation score threshold and the recommendation score corresponding to the object to be recommended, target object data to be displayed to the user is determined, and the target object data is sent to a target terminal device so that the target object data is displayed on a current graphical user interface of the target terminal device.

2. The object recommendation method according to claim 1, characterized in that: The method further comprises: obtaining a user attribute vector, a user historical behavior feature vector, and an attribute vector corresponding to the object to be recommended, and inputting the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended into the trained object recommendation model. Acquire user attribute data, user historical behavior data, and attribute data corresponding to the object to be recommended; The user attribute data, the user historical behavior data, and the attribute data corresponding to the object to be recommended are respectively embedded to obtain the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended.

3. The object recommendation method according to claim 1, characterized in that: The performing session division processing on the user historical behavior feature vector to obtain at least one user historical session vector includes: Performing concatenation processing on the user historical behavior feature vectors to obtain at least one user historical behavior vector sequence; Each of the user history behavior vector sequences is divided and processed based on a preset time node to obtain at least one of the user history session vectors.

4. The object recommendation method according to claim 1, characterized in that: The method of processing each of the user history conversation vectors and the attribute vector corresponding to the to-be-recommended object based on the attention mechanism to obtain a self-attention vector corresponding to each of the user history conversation vectors includes: Performing weight calculation processing on each of the user history conversation vectors and the attribute vector corresponding to the to-be-recommended object to obtain a weight value corresponding to each of the user history conversation vectors; The weight values ​​corresponding to the user history conversation vectors and the user history conversation vectors are subjected to feature fusion processing to obtain self-attention vectors corresponding to the user history conversation vectors.

5. The object recommendation method according to claim 1, characterized in that: The performing of a gating mechanism on each of the user history conversation vectors and the self-attention vector corresponding to each of the user history conversation vectors to obtain a gating vector corresponding to each of the user history conversation vectors includes: Performing nonlinear activation processing on each of the user history conversation vectors and the self-attention vectors corresponding to each of the user history conversation vectors to obtain a retention vector corresponding to each of the user history conversation vectors; The reserved vectors corresponding to the user history session vectors are processed by resetting the gates to obtain the gating vectors corresponding to the user history session vectors.

6. The object recommendation method according to claim 1, characterized in that: Before obtaining the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended, and inputting the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended into the trained object recommendation model, the method further includes: Obtaining user attribute training data, user historical behavior training data, attribute data corresponding to the training object, and real labels corresponding to the training object; The user attribute training data, the user historical behavior training data, and the attribute data corresponding to the training object are respectively embedded to obtain a user attribute training vector, a user historical behavior feature training vector, and an attribute vector corresponding to the training object; Performing session division processing on the user historical behavior feature training vector by using the object recommendation model to obtain at least one user historical session training vector; Based on the attention mechanism, each of the user history conversation training vectors and the attribute vector corresponding to the training object are processed to obtain a self-attention vector corresponding to each of the user history conversation training vectors; Performing gating mechanism processing on each of the user history conversation training vectors and the self-attention vectors corresponding to each of the user history conversation training vectors to obtain a gating vector corresponding to each of the user history conversation training vectors; Fusing the gating vectors corresponding to the user history session training vectors to obtain a user interest training vector; Determining a recommendation score corresponding to the training object according to the user attribute training vector, the user interest training vector, and the attribute vector corresponding to the training object; Determining the loss of the object recommendation model according to the recommendation score corresponding to the training object and the true label corresponding to the training object; In a loop iterative manner, the parameters in the object recommendation model are updated according to the loss corresponding to the object recommendation model.

7. The object recommendation method according to claim 1, characterized in that: The determining, according to the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended, a recommendation score corresponding to the object to be recommended, comprises: Performing splicing processing on the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended to obtain a splicing vector corresponding to the object to be recommended; Flattening the concatenated vector corresponding to the object to be recommended to obtain a low-dimensional vector corresponding to the object to be recommended; Mapping the low-dimensional vector corresponding to the object to be recommended by a multi-layer perceptron to obtain a spatial mapping vector corresponding to the object to be recommended; A recommendation score calculation process is performed on the space mapping vector corresponding to the object to be recommended to obtain a recommendation score corresponding to the object to be recommended.

8. An object recommendation device, characterized in that: include: An acquisition module, used to acquire a user attribute vector, a user historical behavior feature vector, and an attribute vector corresponding to the object to be recommended, and input the user attribute vector, the user historical behavior feature vector, and the attribute vector corresponding to the object to be recommended into the trained object recommendation model; A first processing module, configured to perform session division processing on the user historical behavior feature vector to obtain at least one user historical session vector; A second processing module is used to process each of the user history conversation vectors and the attribute vector corresponding to the object to be recommended based on the attention mechanism to obtain a self-attention vector corresponding to each of the user history conversation vectors; A third processing module is used to perform a gating mechanism processing on each of the user historical conversation vectors and the self-attention vector corresponding to each of the user historical conversation vectors to obtain a gating vector corresponding to each of the user historical conversation vectors; A fourth processing module, configured to fuse the gating vectors corresponding to the user historical session vectors to obtain a user interest vector; A determination module, configured to determine a recommendation score corresponding to the object to be recommended based on the user attribute vector, the user interest vector, and the attribute vector corresponding to the object to be recommended; The fifth processing module is used to determine the target object data to be displayed to the user based on a preset recommendation score threshold and the recommendation score corresponding to the object to be recommended, and send the target object data to the target terminal device so that the target object data is displayed on the current graphical user interface of the target terminal device.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images