Embedding feature extraction method and device for social information
By extracting social information posting information and target users' historical interaction behavior to generate embedding features, the problem of poor quality of social information embedding features is solved, and higher quality and real-time embedding features are achieved, thereby improving the performance of the recommendation system.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- MICRO DREAM TECHTRONIC NETWORK TECH CHINACO
- Filing Date
- 2022-12-23
- Publication Date
- 2026-05-08
AI Technical Summary
In existing technologies, the embedding features of social information are of poor quality and cover a small number of users, resulting in poor performance of recommendation systems.
By acquiring the social information published by the publishers associated with the target user, the first embedding feature of the published information is extracted; the second embedding feature of the target user is determined based on the target user's historical interaction behavior and the first embedding feature; the target social information when the target user engages in real-time interaction behavior is acquired, and a third embedding feature is generated based on the target social information, the first embedding feature, and the second embedding feature.
It improved the quality and real-time performance of social information embedding features, increased the number of users covered, and optimized the performance of the recommendation system.
Smart Images
Figure CN116244499B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of natural language processing technology, and in particular to a method and apparatus for extracting embedding features from social information. Background Technology
[0002] With the continuous development of internet technology, more and more users are consuming content they are interested in through various content platforms. The information explosion brought about by massive amounts of content is constantly driving recommendation systems towards more precise distribution, and the requirements for the accuracy and timeliness of content in recommendation systems are becoming increasingly stringent.
[0003] In some scenarios, embedding user information, user IDs for social messages posted on social products, and other social information has been widely applied in recommendation systems. A common approach to obtaining social information embedding features involves using a pre-trained language model to derive the embedding features based on the type of social message and the information of the poster. However, this method results in embedding features that carry limited information, cover a small number of users, and are of poor quality, leading to poor performance in recommendation systems. Summary of the Invention
[0004] The purpose of this application is to provide a method and apparatus for extracting embedding features from social information, so as to solve the problem of poor quality of embedding features in social information.
[0005] To solve the above-mentioned technical problems, the embodiments of this application are implemented as follows:
[0006] In a first aspect, embodiments of this application provide a method for extracting embedding features from social information, comprising: obtaining posting information corresponding to each social information posted by a publisher associated with a target user; extracting a first embedding feature from each of the posted information; determining a second embedding feature of the target user based on the target user's historical interaction behavior and each of the first embedding features; obtaining target social information when the target user engages in real-time interaction behavior; and obtaining a third embedding feature of the target social information based on the target social information, the first embedding feature, and the second embedding feature.
[0007] Secondly, embodiments of this application provide a device for extracting embedding features of social information. The device includes: an acquisition module for acquiring publishing information corresponding to each piece of social information published by a publisher associated with a target user; an extraction module for extracting a first embedding feature of each piece of publishing information; a determination module for determining a second embedding feature of the target user based on the target user's historical interaction behavior and each of the first embedding features; the acquisition module for acquiring target social information when the target user engages in real-time interaction; and a weighting module for obtaining a third embedding feature of the target social information based on the target social information, the first embedding feature, and the second embedding feature.
[0008] Thirdly, embodiments of this application provide an electronic device, including a processor, a communication interface, a memory, and a communication bus; wherein the processor, the communication interface, and the memory communicate with each other via the bus; the memory is used to store computer programs; and the processor is used to execute the programs stored in the memory to implement the steps as described in the first aspect.
[0009] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method steps as described in the first aspect.
[0010] Fifthly, embodiments of this application provide a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled, and the processor is used to run programs or instructions to implement the method steps as described in the first aspect.
[0011] As can be seen from the technical solutions provided in the embodiments of this application above, by obtaining the publishing information corresponding to each social information published by the publisher associated with the target user; extracting the first embedding feature of each of the published information; determining the second embedding feature of the target user based on the target user's historical interaction behavior and each of the first embedding features; obtaining the target social information when the target user engages in real-time interaction behavior; and obtaining the third embedding feature of the target social information based on the target social information, the first embedding feature, and the second embedding feature, it is possible to ensure that the embedding feature of the generated social information can carry more information about relational behavior from the perspective of the published information and the target user associated with the publisher, thereby increasing the number of users covered. Furthermore, by extracting the embedding feature from the social information of real-time interaction behavior through the first and second embedding features, the quality and real-time performance of the embedding feature are improved, thus optimizing the usage effect for the recommendation system. Attached Figure Description
[0012] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0013] Figure 1 A flowchart illustrating the method for extracting embedding features of social information provided in this application embodiment;
[0014] Figure 2 The graph showing the change of the time decay function provided in the embodiments of this application;
[0015] Figure 3 A schematic diagram of the functional modules of the social information embedding feature extraction device provided in the embodiments of this application;
[0016] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0017] The purpose of this application is to provide a method and apparatus for extracting embedding features of social information, thereby improving the quality of embedding features of social information.
[0018] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.
[0019] To address the above technical problems, this application provides a method and apparatus for extracting embedding features of social information. The following is a detailed description of the method and apparatus for extracting embedding features of social information provided in this application, with reference to the accompanying drawings.
[0020] like Figure 1 As shown, this application embodiment provides a method for extracting embedding features of social information. The execution subject of this method can be a server, which can be a standalone server or a server cluster composed of multiple servers. Specifically, this method for extracting embedding features of social information may include the following steps S101-S109:
[0021] In step S101, the posting information corresponding to each social information posted by the publisher associated with the target user is obtained.
[0022] Specifically, social information can be information posted by users through social media platforms such as Weibo, WeChat, and Voice. Social information includes, but is not limited to, text, images, and combinations of text and images. Social information can be posted by a user, who can be the user who posts the social information. The posted information contains the user's personal information, including but not limited to the posting account (poster's ID) and tags associated with the posted information. Target users refer to users who have historical interaction with the posted information. Types of historical interaction include, but are not limited to, liking, forwarding, commenting, following, and saving.
[0023] In step S103, the first embedding feature of each published information is extracted.
[0024] Specifically, the first embedding feature is a low-dimensional vector feature used to represent the published information. Since the number of publishers is relatively small compared to the entire user group, and the tag information is relatively rich and accurate, the first embedding feature can be extracted using the publisher's ID information and social tag information. The BERT language model is a widely used language model in the field of natural language processing. By inputting pre-processed sentences into the BERT language model, embedding features of sentences and words can be trained.
[0025] In one possible implementation, the published information includes: the publisher's ID information and the social information's tag information. Extracting the first embedding feature of each published information includes: composing a sentence from the publisher's ID information and tag information, inputting the sentence into the BERT language model for feature extraction processing, and obtaining the first embedding feature of the published information. The embedding feature includes tag embedding feature and publisher embedding feature, and the tag embedding feature and publisher embedding feature have a corresponding relationship.
[0026] Specifically, the tag information of social information can be multi-level tags manually labeled, including the type of social information, the circle to which the social information belongs, etc. The publisher's ID information refers to the publisher's account information or the publisher's registration number, etc. When extracting the first embedding feature of the published information, the publisher's ID information and tag information can be combined into a statement, and the statement can be input into the BERT language model for feature extraction processing. The BERT language model outputs the first embedding feature of the published information, which carries the tag embedding feature of the tag information and the publisher embedding feature of the publisher's ID information.
[0027] In step S105, the second embedding feature of the target user is determined based on the target user's historical interaction behavior and each first embedding feature.
[0028] Specifically, the types of historical interaction behaviors include, but are not limited to, liking, forwarding, commenting, following, and collecting. The embedding characteristics of the target user are determined by the target user's historical interaction behaviors with the information published by the publisher. Specifically, it can be obtained by weighting the various types of behaviors of the target user with the information published by the publisher.
[0029] In one possible implementation, determining the second embedding feature of the target user based on the target user's historical interaction behavior and the first embedding feature includes: obtaining a first number of historical interaction behaviors of the target user, a second number of publishers associated with the target user, and the first embedding feature of each published message; determining the weights of different behavior types in the target user's historical interaction behaviors; and determining the second embedding feature based on the first number, the second number, the first embedding feature of each publisher, and the weights of different behavior types in the historical interaction behaviors.
[0030] Specifically, the first quantity refers to the number of historical interactions between the target user and the publisher, specifically the number of interactions between the target user and the publisher within a predetermined time period. The second quantity refers to the number of interactions between the target user and the publisher, specifically the number of interactions between the target user and the publisher within a predetermined time period. The weights assigned to different types of interactions by the target user on social media messages can vary; for example, forwarding may have the highest weight, followed by commenting, and likes may have less weight than comments. The second embedding feature, determined using the first quantity, the second quantity, the first embedding feature of each posted message, and the weights of different behavior types in historical interactions, can be calculated using the following formula:
[0031]
[0032] Where userEmb is the second embedding feature, m is the first quantity, n is the second quantity, and b j bloggerEmb assigns weights to different behavior types within the target user's historical interaction history. i This is the first embedding feature.
[0033] In this case, due to the large number of target users, the limited number of individual user behaviors, and the sparse behavioral data, the embedding features obtained solely using language models are ineffective. Because of the long-tail effect in user historical interaction behavior—that is, a small number of publishers occupy a large amount of exposure, and there is massive interaction between users and publishers—an attention mechanism can be used to weight the publisher's embedding features to obtain the target user's embedding features. The attention mechanism originates from research on human vision, focusing on key inputs and allocating more limited resources to important parts. In this embodiment, since user interaction behaviors include forwarding, commenting, liking, following, and collecting, and different behaviors express different levels of user preference, statistical analysis of past user behaviors is performed. Combined with business objectives, different weights are assigned to different types of interaction behaviors, with core behaviors such as forwarding receiving greater weight. All weights are multiplied by the corresponding publisher's embedding and then weighted and averaged to implement the attention mechanism and ultimately obtain the user's embedding.
[0034] To ensure user weights are as clear as possible, the time factor is considered in user historical interaction behavior. A time decay coefficient is multiplied by the initial weight, and this coefficient is used to fit the decline curve of user interest. The time decay coefficient statistics reference the historical behavior of the vast majority of users, employing methods such as... Figure 2 The S-shaped curve shown simulates the function of user interest decline towards the publisher, with a gentle decline at both ends and a larger decline in the middle. By introducing a time decay coefficient, the accuracy and effectiveness of the target user's second embedding feature are significantly improved. Based on this, this application provides a possible implementation method for determining the weights of different behavior types in the target user's historical interaction behavior, including: assigning different initial weights to different behavior types in historical interaction behavior according to the priority of the behavior type based on an attention mechanism, wherein the priority of the behavior type is proportional to the initial weight;
[0035] By multiplying the initial weights by a time decay coefficient, we obtain the weights of different behavior types in the target user's historical interaction behavior. The time decay coefficient is inversely proportional to the time difference between the user's interaction behavior and the current time.
[0036] Specifically, after multiplying the initial weights by introducing a time decay function, the second embedding feature is calculated using the following formula:
[0037]
[0038] Where userEmb is the second embedding feature, m is the first quantity, n is the second quantity, a is the time difference between the user's interaction and the current time, and b is the second embedding feature.j bloggerEmb assigns weights to different behavior types within the target user's historical interaction history. i This is the first embedding feature.
[0039] In step S107, target social information is obtained when the target user engages in real-time interactive behavior.
[0040] Specifically, real-time user interaction refers to the user's real-time interaction at the current time, and target social information refers to the social information generated by the user's real-time interaction at the current time. By using the user's real-time interaction, the embedding features of the target social information with real-time interaction can be obtained in real time by weighting the user's real-time interaction, the first embedding feature of the published information, and the second embedding feature of the target user, achieving second-level updates of the embedding features.
[0041] In step S109, the third embedding feature of the target social information is obtained based on the target social information, the first embedding feature, and the second embedding feature.
[0042] Specifically, for users' real-time interactive behavior, the third embedding feature of the target social information can be determined according to whether the target social information is associated with the publisher.
[0043] The third embedding feature, obtained by weighting the first and second embedding features based on the target social information, includes:
[0044] If the real-time interactive behavior corresponding to the target social information is not the first occurrence, the third embedding feature of the target social information is obtained by weighting the fourth embedding feature corresponding to the published information corresponding to the target social information, the second embedding feature of the target user corresponding to the target social information, and the third number of users interacting with the target social information.
[0045] Specifically, the first and second embedding features can be synchronized to the Redis database for pre-storage. When there is a correspondence between the first embedding feature and the publisher of the target social information, if the real-time interactive behavior corresponding to the target social information is not the first occurrence, or if the embedding feature of the target social information that interacts with the current user in real time exists in the Redis database, then a new embedding feature is added based on the fourth embedding feature corresponding to the published information corresponding to the target social information, and the embedding feature of the target social information (the fourth embedding feature) pre-stored in the Redis database is overwritten.
[0046] Specifically, the following formula can be used for calculation:
[0047]
[0048] Where blogEmb is the third embedding feature, n1 is the third quantity, blogEmb0 is the fourth embedding feature, a0 is a random number function that randomly generates 0 and 1 to achieve random downsampling, and userEmb... i This is the second embedding feature.
[0049] When the real-time interactive behavior corresponding to the target social information occurs for the first time, the third number of users interacting with the target social information, the first embedding feature, and the second embedding feature are weighted to obtain the third embedding feature of the target social information.
[0050] If the Redis database does not contain the embedding feature of the target social information that is currently interacting with the user in real time, that is, when the real-time interaction behavior corresponding to the target social information occurs for the first time, the third embedding feature of the target social information is calculated using the following formula:
[0051]
[0052] Where blogEmb is the third embedding feature, bloggerEmb is the first embedding feature, n1 is the third quantity, a0 is the random number function, and userEmb... i This is the second embedding feature.
[0053] The technical solution disclosed in this application involves obtaining the publishing information corresponding to each social information published by a publisher associated with a target user; extracting the first embedding feature of each published information; determining the second embedding feature of the target user based on the target user's historical interaction behavior and each of the first embedding features; obtaining the target social information when the target user engages in real-time interaction behavior; and obtaining the third embedding feature of the target social information based on the target social information, the first embedding feature, and the second embedding feature. This approach ensures that the generated social information's embedding feature carries more information about relational behaviors from the perspective of the published information and the target users associated with the publisher, increasing the number of users covered. Furthermore, by extracting embedding features from the social information of real-time interaction behavior using the first and second embedding features, the quality and real-time performance of the embedding features are improved, thus optimizing the usage effect for the recommendation system.
[0054] Corresponding to the social information embedding feature extraction method provided in the above embodiments, based on the same technical concept, this application also provides a social information embedding feature extraction device. Figure 3 This is a schematic diagram of the module composition of the social information embedding feature extraction device provided in the embodiments of this application. This social information embedding feature extraction device is used to execute the social information embedding feature extraction method described in the above embodiments, such as... Figure 3 As shown, the social information embedding feature extraction device 300 includes: an acquisition module 301, used to acquire the publishing information corresponding to each social information published by the publisher associated with the target user; an extraction module 302, used to extract the first embedding feature of each publishing information; a determination module 303, used to determine the second embedding feature of the target user based on the target user's historical interaction behavior and the first embedding feature; an acquisition module 301, used to acquire the target social information when the target user engages in real-time interaction behavior; and a weighting module 304, used to obtain the third embedding feature of the target social information based on the target social information, the first embedding feature, and the second embedding feature.
[0055] The technical solution disclosed in this application involves obtaining the publishing information corresponding to each social information published by a publisher associated with a target user; extracting the first embedding feature of each published information; determining the second embedding feature of the target user based on the target user's historical interaction behavior and each of the first embedding features; obtaining the target social information when the target user engages in real-time interaction behavior; and obtaining the third embedding feature of the target social information based on the target social information, the first embedding feature, and the second embedding feature. This approach ensures that the generated social information's embedding feature carries more information about relational behaviors from the perspective of the published information and the target users associated with the publisher, increasing the number of users covered. Furthermore, by extracting embedding features from the social information of real-time interaction behavior using the first and second embedding features, the quality and real-time performance of the embedding features are improved, thus optimizing the usage effect for the recommendation system.
[0056] In one possible implementation, the published information includes: the publisher's ID information and the social information's tag information. The extraction module 302 is also used to combine the publisher's ID information and tag information into a statement, input the statement into the BERT language model for feature extraction processing, and obtain the first embedding feature of the published information. The first embedding feature includes the tag embedding feature and the publisher embedding feature, and the tag embedding feature and the publisher embedding feature have a corresponding relationship.
[0057] In one possible implementation, the determining module 303 is further configured to obtain a first number of historical interaction behaviors of the target user, a second number of publishers associated with the target user, and a first embedding feature of each published information; determine the weights of different behavior types in the historical interaction behaviors of the target user; and determine a second embedding feature based on the first number, the second number, the first embedding feature of each published information, and the weights of different behavior types in the historical interaction behaviors.
[0058] In one possible implementation, the determining module 303 is further configured to assign different initial weights to different behavior types in historical interaction behaviors based on the priority of behavior types according to the attention mechanism, wherein the priority of behavior types is directly proportional to the initial weights; and multiply the initial weights by a time decay coefficient to obtain the weights of different behavior types in the target user's historical interaction behaviors, wherein the time decay coefficient is inversely proportional to the time difference between the user's interaction behavior and the current time.
[0059] In one possible implementation, the second embedding feature is calculated using the following formula:
[0060]
[0061] Where userEmb is the second embedding feature, m is the first quantity, n is the second quantity, a is the time difference between the user's interaction and the current time, and b is the second embedding feature. j bloggerEmb assigns weights to different behavior types within the target user's historical interaction history. i This is the first embedding feature.
[0062] In one possible implementation, the weighting module 304 is further configured to, when the real-time interactive behavior corresponding to the target social information is not the first occurrence, weight the fourth embedding feature corresponding to the published information corresponding to the target social information, the second embedding feature of the target user corresponding to the target social information, and the third number of users interacting with the target social information to obtain the third embedding feature of the target social information; and when the real-time interactive behavior corresponding to the target social information is the first occurrence, weight the third number of users interacting with the target social information, the first embedding feature, and the second embedding feature to obtain the third embedding feature of the target social information.
[0063] In one possible implementation, when the real-time interactive behavior corresponding to the target social information is not the first occurrence, the third embedding feature of the target social information is calculated using the following formula:
[0064]
[0065] Where blogEmb is the third embedding feature, n1 is the third quantity, blogEmb0 is the fourth embedding feature, a0 is the random number function, and userEmb... i This is the second embedding feature.
[0066] When the real-time interactive behavior corresponding to the target social information appears for the first time, the third embedding feature of the target social information is calculated using the following formula:
[0067]
[0068] Where blogEmb is the third embedding feature, bloggerEmb is the first embedding feature, n1 is the third quantity, a0 is the random number function, and userEmb... i This is the second embedding feature.
[0069] The social information embedding feature extraction device provided in this application embodiment can realize the various processes in the embodiments corresponding to the above-mentioned social information embedding feature extraction method. To avoid repetition, it will not be described again here.
[0070] It should be noted that the social information embedding feature extraction device provided in this application embodiment and the social information embedding feature extraction method provided in this application embodiment are based on the same inventive concept and have the same technical effect. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned social information embedding feature extraction method, and the repeated parts will not be described again.
[0071] Corresponding to the above-described method for extracting social information embedding features, based on the same technical concept, this application also provides an electronic device for performing the above-described method for extracting social information embedding features. Figure 4 To illustrate the structure of an electronic device according to various embodiments of the present invention, as shown in the schematic diagram... Figure 4 As shown. Electronic devices can vary considerably due to differences in configuration or performance, and may include one or more processors 401 and memory 402. Memory 402 may store one or more application programs or data. Memory 402 may be temporary or persistent storage. The application programs stored in memory 402 may include one or more modules (not shown), each module may include a series of computer-executable instructions for the electronic device. Furthermore, processor 401 may be configured to communicate with memory 402 and execute the series of computer-executable instructions in memory 402 on the electronic device. The electronic device may also include one or more power supplies 403, one or more wired or wireless network interfaces 404, one or more input / output interfaces 405, and one or more keyboards 406.
[0072] In this embodiment, the electronic device includes a processor, a communication interface, a memory, and a communication bus; wherein, the processor, the communication interface, and the memory communicate with each other through the bus; the memory is used to store computer programs; the processor is used to execute the programs stored in the memory to implement the steps described in the above method embodiment.
[0073] It should be noted that the electronic device provided in this application embodiment and the social information embedding feature extraction method provided in this application embodiment are based on the same inventive concept and have the same technical effect. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned social information embedding feature extraction method, and the repeated parts will not be described again.
[0074] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps described in the above method embodiments.
[0075] It should be noted that the computer-readable storage medium provided in this application embodiment and the social information embedding feature extraction method provided in this application embodiment are based on the same inventive concept and have the same technical effect. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned social information embedding feature extraction method, and the repeated parts will not be described again.
[0076] In a specific embodiment, this application provides a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled, and the processor is used to run programs or instructions to implement the steps described in the above method embodiments.
[0077] It should be noted that the chip provided in this application embodiment and the social information embedding feature extraction method provided in this application embodiment are based on the same inventive concept and have the same technical effect. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned social information embedding feature extraction method, and the repeated parts will not be described again.
[0078] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0079] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0080] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0081] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0082] In a typical configuration, an electronic device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0083] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.
[0084] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.
[0085] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0086] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0087] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.
Claims
1. A method for extracting embedding features from social information, characterized in that, include: Obtain the posting information corresponding to each social information posted by the publisher associated with the target user; Extract the first embedding feature of each of the published information; The second embedding feature of the target user is determined based on the target user's historical interaction behavior and each of the first embedding features; the target social information of the target user when real-time interaction behavior occurs is obtained; Based on the target social information, the first embedding feature, and the second embedding feature, a third embedding feature of the target social information is obtained; The step of obtaining the third embedding feature of the target social information based on the target social information, the first embedding feature, and the second embedding feature includes: If the real-time interactive behavior corresponding to the target social information is not the first occurrence, the fourth embedding feature corresponding to the published information corresponding to the target social information, the second embedding feature of the target user corresponding to the target social information, and the third number of users interacting with the target social information are weighted to obtain the third embedding feature of the target social information. When a real-time interactive behavior corresponding to the target social information occurs for the first time, the third number of users interacting with the target social information, the first embedding feature, and the second embedding feature are weighted to obtain the third embedding feature of the target social information.
2. The method for extracting embedding features of social information according to claim 1, characterized in that, The published information includes: the ID information of the publisher of the social information and the tag information of the social information; the extraction of the first embedding feature of each published information includes: The publisher's ID information and the tag information are combined to form a statement. The statement is then input into the BERT language model for feature extraction processing to obtain the first embedding feature of the published information. The first embedding feature includes tag embedding features and publisher embedding features, and the tag embedding features and publisher embedding features have a corresponding relationship.
3. The method for extracting embedding features of social information according to claim 1, characterized in that, The step of determining the second embedding feature of the target user based on the target user's historical interaction behavior and each of the first embedding features includes: Obtain a first number of the target user's historical interaction behaviors, a second number of publishers associated with the target user, and a first embedding feature for each of the published messages; Determine the weights of different behavior types among the target user's historical interaction behaviors; The second embedding feature is determined based on the first quantity, the second quantity, the first embedding feature of each of the published information, and the weights of different behavior types in the historical interaction behavior.
4. The method for extracting embedding features of social information according to claim 3, characterized in that, The determination of the weights for different behavior types among the target user's historical interaction behaviors includes: Based on the attention mechanism, different initial weights are assigned to different behavior types in the historical interaction behavior according to the priority of the behavior type, wherein the priority of the behavior type is proportional to the initial weight; The initial weights are multiplied by a time decay coefficient to obtain the weights of different behavior types in the target user's historical interaction behavior, wherein the time decay coefficient is inversely proportional to the time difference between the user's interaction behavior and the current time.
5. The method for extracting embedding features of social information according to claim 4, characterized in that, The second embedding feature is calculated using the following formula: in, This is the second embedding feature. As the first quantity, For the second quantity, The time difference between the user's interaction and the current time. Weights are assigned to different behavior types among the target user's historical interaction behaviors. This is the first embedding feature.
6. The method for extracting embedding features of social information according to claim 1, characterized in that, If the real-time interactive behavior corresponding to the target social information is not the first occurrence, the third embedding feature of the target social information is calculated using the following formula: in, This is the third embedding feature. As the third quantity, This is the fourth embedding feature. For random number functions, This is the second embedding feature; When the real-time interactive behavior corresponding to the target social information occurs for the first time, the third embedding feature of the target social information is calculated using the following formula: in, This is the third embedding feature. This is the first embedding feature. As the third quantity, For random number functions, This is the second embedding feature.
7. A device for extracting embedding features from social information, characterized in that, include: The acquisition module is used to acquire the posting information corresponding to each social information posted by the publisher associated with the target user; The extraction module is used to extract the first embedding feature of each of the published information items; The determining module is configured to determine the second embedding feature of the target user based on the target user's historical interaction behavior and each of the first embedding features; The acquisition module is used to acquire target social information when the target user engages in real-time interactive behavior; The weighting module is used to obtain a third embedding feature of the target social information based on the target social information, the first embedding feature, and the second embedding feature; The weighting module is further configured to, when the real-time interactive behavior corresponding to the target social information is not the first occurrence, weight the fourth embedding feature corresponding to the published information corresponding to the target social information, the second embedding feature of the target user corresponding to the target social information, and the third number of users interacting with the target social information to obtain the third embedding feature of the target social information. When a real-time interactive behavior corresponding to the target social information occurs for the first time, the third number of users interacting with the target social information, the first embedding feature, and the second embedding feature are weighted to obtain the third embedding feature of the target social information.
8. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the embedding feature extraction method for social information as described in any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is able to perform the embedding feature extraction method for social information as described in any one of claims 1 to 6.