CN-116244499-B - Embedding feature extraction method and embedding feature extraction device of social information

CN116244499BCN 116244499 BCN116244499 BCN 116244499BCN-116244499-B

Abstract

The embodiment of the application provides a embedding feature extraction method and device of social information, and the method and device comprise the steps of obtaining posting information corresponding to each social information posted by a publisher associated with a target user, extracting first embedding features of each posting information, determining second embedding features of the target user according to historical interaction behaviors of the target user and first embedding features, obtaining target social information when the target user has real-time interaction behaviors, and obtaining third embedding features of the target social information according to the target social information, the first embedding features and the second embedding features.

Inventors

ZHANG KAI
LIU DAOGUANG
LIU BO

Assignees

微梦创科网络科技（中国）有限公司

Dates

Publication Date: 20260508
Application Date: 20221223

Claims (9)

1. A method for extracting embedding features of social information, comprising: Acquiring release information corresponding to each social information released by a publisher associated with a target user; Extracting a first embedding characteristic of each piece of release information; Determining a second embedding characteristic of the target user according to the historical interaction behavior of the target user and each first embedding characteristic; obtaining a third embedding feature of the target social information according to the target social information, the first embedding feature and the second embedding feature; the obtaining the third embedding feature of the target social information according to the target social information, the first embedding feature and the second embedding feature includes: Weighting a fourth embedding feature corresponding to posting information corresponding to the target social information, a second embedding feature corresponding to the target user corresponding to the target social information and a third number of users who interacted with the target social information to obtain a third embedding feature of the target social information when the real-time interaction behavior corresponding to the target social information does not occur for the first time; And under the condition that the real-time interaction behavior corresponding to the target social information appears for the first time, weighting the third number of users interacting with the target social information, the first embedding features and the second embedding features to obtain third embedding features of the target social information.
2. The method for extracting embedding features of social information according to claim 1, wherein the posting information includes ID information of a posting of the social information and tag information of the social information, and extracting the first embedding feature of each posting information includes: And forming sentences by the ID information of the publisher and the tag information, inputting the sentences into a Bert language model for feature extraction processing to obtain a first embedding feature of the published information, wherein the first embedding feature comprises a tag embedding feature and a publisher embedding feature, and the tag embedding feature and the publisher embedding feature have a corresponding relationship.
3. The method of claim 1, wherein determining the second embedding feature of the target user based on the historical interaction behavior of the target user and each of the first embedding features comprises: Acquiring 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 characteristic of each piece of published information; determining weights of different behavior types in the historical interaction behaviors of the target user; and determining the second embedding characteristic according to the first quantity, the second quantity, the first embedding characteristic of each piece of release information and weights of different types of behaviors in the historical interaction behaviors.
4. The method of claim 3, wherein determining weights for different types of behavior in the historical interaction behavior of the target user comprises: different initial weights are given to different behavior types in the historical interaction behavior based on an attention mechanism according to the priority of the behavior types, wherein the priority of the behavior types is in direct proportion to the initial weights; And multiplying the initial weight by a time attenuation coefficient to obtain weights of different behavior types in the historical interaction behavior of the target user, wherein the time attenuation coefficient is inversely proportional to the time difference between the interaction behavior of the user and the current time.
5. The method of claim 4, wherein the second embedding feature is calculated using the formula: Wherein, the As a feature of the second embedding-feature, For the first number of times, In the case of the second number of the first number, For the user to have a time difference between the interaction behavior and the current time, Weights of different types of behavior in the historical interaction behavior occur for the target user, Is a first embedding feature.
6. The method of claim 1, wherein in the event that the real-time interaction behavior corresponding to the target social information is not the first occurrence, calculating a third embedding feature of the target social information using the following formula: Wherein, the In the case of the third embedding-feature, For the third number of times, In order to provide the fourth embedding feature, As a function of the random number(s), Is the second embedding features; Under the condition that the real-time interaction behavior corresponding to the target social information appears for the first time, calculating a third embedding characteristic of the target social information by adopting the following formula: Wherein, the In the case of the third embedding-feature, In the first embedding-feature set, For the third number of times, As a function of the random number(s), Is the second embedding features.
7. A embedding feature extraction device for social information, comprising: the acquisition module is used for acquiring release information corresponding to each social information released by the release person associated with the target user; the extraction module is used for extracting the first embedding features of each piece of release information; The determining module is used for determining second embedding characteristics of the target user according to the historical interaction behaviors of the target user and the first embedding characteristics; the acquisition module is used for acquiring target social information when the target user generates real-time interaction behavior; the weighting module is used for obtaining a third embedding feature of the target social information according to the target social information, the first embedding feature and the second embedding feature; The weighting module is further configured to weight, when the real-time interaction behavior corresponding to the target social information does not occur for the first time, a fourth embedding feature corresponding to posting information corresponding to the target social information, a second embedding feature of a target user corresponding to the target social information, and a third number of users interacting with the target social information, to obtain a third embedding feature of the target social information; And under the condition that the real-time interaction behavior corresponding to the target social information appears for the first time, weighting the third number of users interacting with the target social information, the first embedding features and the second embedding features to obtain third embedding features of the target social information.
8. An electronic device, comprising: A processor; a memory for storing the processor-executable instructions; Wherein the processor is configured to execute the instructions to implement the embedding feature extraction method of social information as claimed in any one of claims 1 to 6.
9. A computer readable storage medium, wherein instructions in the storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the embedding feature extraction method of social information as claimed in any one of claims 1 to 6.

Description

Embedding feature extraction method and embedding feature extraction device of social information Technical Field The invention relates to the technical field of natural language processing, in particular to a embedding feature extraction method and device of social information. Background With the continuous development of internet technology, more and more users consume content of interest by themselves through various content platforms. Information explosion brought by mass content continuously pushes a recommendation system to precisely distribute, and the recommendation system has stronger and stronger requirements on the accuracy and timeliness of the content. In some scenarios, social information is widely used in recommendation systems by including users, ID information of the users who post social information through social products, and social information embedding. The adopted main implementation means for acquiring embedding features of social information is that embedding features of social information are obtained through a pre-trained language model aiming at the types of the social information and the information of a publisher for publishing the social information, so that the finally obtained embedding features carry less information, the number of covered users is less, the quality of the obtained embedding features is poor, and the use effect is poor for a recommendation system. Disclosure of Invention The embodiment of the application aims to provide a embedding feature extraction method and device of social information, which are used for solving the problem of poor quality of embedding features of the social information. In order to solve the technical problems, the embodiment of the application is realized as follows: In a first aspect, an embodiment of the present application provides a method for extracting embedding features of social information, including obtaining posting information corresponding to each social information posted by a posting associated with a target user, extracting a first embedding feature of each posting information, determining a second embedding feature of the target user according to a historical interaction behavior of the target user and each first embedding feature, obtaining target social information when the target user has a real-time interaction behavior, and obtaining a third embedding feature of the target social information according to the target social information, the first embedding feature and the second embedding feature. In a second aspect, an embodiment of the application provides a embedding feature extraction device of social information, wherein the embedding feature extraction device of social information comprises an acquisition module, an extraction module, a determination module and a weighting module, wherein the acquisition module is used for acquiring posting information corresponding to each social information posted by a posting associated with a target user, the extraction module is used for extracting a first embedding feature of each posting information, the determination module is used for determining a second embedding feature of the target user according to the historical interaction behavior of the target user and each first embedding feature, the acquisition module is used for acquiring target social information when the target user has real-time interaction behavior, and the weighting module is used for obtaining a third embedding feature of the target social information according to the target social information, the first embedding feature and the second embedding feature. In a third aspect, an embodiment of the present application provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, where the processor, the communication interface, and the memory complete communication with each other through the bus, the memory is configured to store a computer program, and the processor is configured to execute the program stored in the memory, to implement the steps as in the first aspect. In a fourth aspect, embodiments of the present application provide a computer readable storage medium having a computer program stored thereon, which when executed by a processor, implements the method steps as in the first aspect. In a fifth aspect, an embodiment of the present application provides a chip, the chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being configured to execute programs or instructions to implement the method steps as in the first aspect. According to the technical scheme provided by the embodiment of the application, through obtaining the posting information corresponding to each social information posted by the posting associated with the target user, extracting the first embedding characteristic of each posting information, determining the second embedding characterist