CN-121980075-A - Data recommendation method, device, equipment and medium

CN121980075ACN 121980075 ACN121980075 ACN 121980075ACN-121980075-A

Abstract

The present application relates to the field of big data analysis technologies, and in particular, to a data recommendation method, device, equipment, and medium. According to the method, fusion is carried out according to interest vectors corresponding to four pieces of different scale information to obtain fusion vectors, current candidate data are screened according to the fusion vectors, the screened data are obtained to be four-scale fusion of recommendation results through labels, browsing histories, feedback behaviors and searching intentions, new and old users can recommend the data based on the method, applicability of the model is improved, and based on diversity analysis of the different scale information, data recommendation flow is optimized to form corresponding accurate interest information, and accurate recommendation results can be obtained.

Inventors

Gong Oubo
YU XIAOTIAN
LI AIJUN

Assignees

深圳云天励飞技术股份有限公司

Dates

Publication Date: 20260505
Application Date: 20251215

Claims (10)

1. A data recommendation method, comprising: Acquiring first scale information, second scale information, third scale information and fourth scale information, wherein the first scale information comprises labels selected by a user, the second scale information comprises historical browsing data, the third scale information comprises feedback behaviors of data in a historical browsing process, and the fourth scale comprises search words of the user in a historical use process; Acquiring current candidate data, determining a first interest vector according to the tag in the first scale information and the current candidate data, and determining a fourth interest vector according to the search word in the fourth scale information and the current candidate data; Performing feature analysis on the historical browsing data in the second scale information to obtain a second interest vector, and performing feature analysis on the feedback behavior in the third scale information to obtain a third interest vector; And obtaining the weight corresponding to each piece of scale information, carrying out weighted fusion on the first interest vector, the second interest vector, the third interest vector and the fourth interest vector according to the weight of each piece of scale information to obtain a fusion vector, and screening the current candidate data according to the fusion vector to obtain screening data as a recommendation result.
2. The data recommendation method according to claim 1, wherein the obtaining weights corresponding to each scale information includes: calculating the confidence coefficient of the first scale information, the second scale information, the third scale information and the fourth scale information based on a preset confidence coefficient condition to obtain the confidence coefficient corresponding to each scale information; Based on a preset dynamic adjustment mechanism, the weight of the corresponding scale information is adjusted by combining the confidence coefficient corresponding to each scale information, so that the weight corresponding to each scale information is obtained, wherein if the confidence coefficient of one scale information is lower, the weight of other scale information is higher, otherwise, the weight of other scale information is lower.
3. The data recommendation method according to claim 1, wherein the determining a first interest vector according to the tag in the first scale information and the current candidate data comprises: Performing vector coding on the label in the first scale information to obtain a label matching vector, and matching the label vector with the current candidate data to obtain first matching data; carrying out semantic analysis on the labels in the first scale information to obtain label semantic vectors, carrying out semantic analysis on the current candidate data to obtain candidate data semantic vectors, and obtaining second matching data according to the similarity of the label semantic vectors and the candidate data semantic vectors; and carrying out weighted fusion on the semantic vector of the first matching data and the semantic vector of the second matching data to obtain a first interest vector.
4. The data recommendation method according to claim 1, wherein the determining a fourth interest vector from the search term in the fourth scale information and the current candidate data comprises: Vector encoding is carried out on the search words in the fourth scale information to obtain search word vectors; vector encoding is carried out on the current candidate data to obtain candidate data encoding vectors; and calculating the similarity according to the search word vector and the candidate data coding vector, and carrying out weighted summation on the candidate data coding vector with the similarity meeting the preset condition to obtain a fourth interest vector.
5. The data recommendation method according to claim 1, wherein the performing feature analysis on the historical browsing data in the second scale information to obtain a second interest vector includes: according to the ascending order of the time stamps, arranging each data in the historical browsing data in the second scale information to obtain a browsing sequence; Extracting target texts from each data in the browsing sequence, carrying out semantic analysis on each target text to obtain a first data semantic vector, and traversing the browsing sequence to obtain a semantic vector sequence; Acquiring browsing duration of each data, analyzing the semantic vector sequence according to the browsing duration, and determining the weight of each first data semantic vector, wherein the weight is higher if the browsing duration of the corresponding data is longer, and the weight is higher if the ordering of the corresponding data is more rearward; and carrying out weighted summation on all the data semantic vectors according to the weight of each first data semantic vector to obtain a second interest vector.
6. The data recommendation method according to any one of claims 1 to 5, wherein the performing feature analysis on the feedback behavior in the third scale information to obtain a third interest vector includes: Classifying the feedback behaviors in the third scale information to obtain data corresponding to the front behaviors; And respectively carrying out semantic analysis on the data corresponding to the front behaviors to obtain second data semantic vectors, and averaging all the second data semantic vectors to obtain a third interest vector.
7. The data recommendation method according to claim 6, wherein when classifying the feedback behavior in the third scale information to obtain data corresponding to the front behavior, further comprising: classifying the feedback behaviors in the third scale information to obtain data corresponding to negative behaviors; Respectively carrying out semantic analysis on the data corresponding to the negative behaviors to obtain second data semantic vectors, and averaging all the second data semantic vectors to obtain negative vectors; the step of screening the current candidate data according to the fusion vector to obtain screening data as a recommended result, including: Calculating a first matching score of each candidate data according to the fusion vector and the current candidate data, and calculating a second matching score of each candidate data according to the negative vector and the current candidate data; Subtracting the corresponding second matching score from the first matching score of the candidate data to obtain a final matching score for any candidate data, and traversing all candidate data to obtain the final matching score of all candidate data; and screening the current candidate data according to the final matching scores of all the candidate data to obtain screening data as a recommended result.
8. A data recommendation device, comprising: The information acquisition module is used for acquiring first scale information, second scale information, third scale information and fourth scale information, wherein the first scale information comprises a label selected by a user, the second scale information comprises historical browsing data, the third scale information comprises feedback behaviors of data in a historical browsing process, and the fourth scale comprises search words of the user in a historical use process; the first interest vector module is used for acquiring current candidate data, determining a first interest vector according to the tag in the first scale information and the current candidate data, and determining a fourth interest vector according to the search word in the fourth scale information and the current candidate data; The second interest vector module is used for carrying out feature analysis on the historical browsing data in the second scale information to obtain a second interest vector, and carrying out feature analysis on the feedback behavior in the third scale information to obtain a third interest vector; The data recommendation module is used for acquiring the weight corresponding to each piece of scale information, carrying out weighted fusion on the first interest vector, the second interest vector, the third interest vector and the fourth interest vector according to the weight of each piece of scale information to obtain a fusion vector, and screening the current candidate data according to the fusion vector to obtain screening data as recommendation results.
9. A computer device, characterized in that it comprises a processor, a memory and a computer program stored in the memory and executable on the processor, which processor implements the data recommendation method according to any of claims 1 to 7 when executing the computer program.
10. A computer readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the data recommendation method according to any one of claims 1 to 7.

Description

Data recommendation method, device, equipment and medium Technical Field The present application relates to the field of big data analysis technologies, and in particular, to a data recommendation method, device, equipment, and medium. Background With the development of big data analysis technology, personalized recommendation has become an indispensable means for service users, for example, current mainstream news recommendation systems commonly construct personalized recommendation systems Based on collaborative Filtering (Collaborative Filtering, CF), content-Based Filtering (CBF) or deep neural Network models (e.g., DEEP INTEREST Network, DIN and Multi-Modal Mixture of Experts, MMOE). However, 1) cold start seriously deviates from individual interests in the existing recommendation process, 2) multi-source user behavior signals cannot form unified preference characterization, recommendation is not facilitated, 3) search behaviors of users are not reasonably used, so that user intention identification is inaccurate, 4) high-quality mass content can be subjected to low exposure in some cases, use of the users is not facilitated, and 5) users in cold start and hot start respectively use different recommendation models, so that system redundancy and high maintenance cost are caused. As can be seen, with the increasing diversification of user demands and the increasing complexity of news content ecology, the traditional recommendation paradigm exposes systematic defects in real business scenes, and is difficult to support high-quality sustainable personalized recommendation services. Therefore, how to optimize the data recommendation process to improve the accuracy of recommendation is a problem to be solved. Disclosure of Invention In view of this, the embodiments of the present application provide a data recommendation method, apparatus, device, and medium, so as to solve the problem of how to optimize a data recommendation flow and improve the accuracy of recommendation. In a first aspect, an embodiment of the present application provides a data recommendation method, including: Acquiring first scale information, second scale information, third scale information and fourth scale information, wherein the first scale information comprises labels selected by a user, the second scale information comprises historical browsing data, the third scale information comprises feedback behaviors of data in a historical browsing process, and the fourth scale comprises search words of the user in a historical use process; Acquiring current candidate data, determining a first interest vector according to the tag in the first scale information and the current candidate data, and determining a fourth interest vector according to the search word in the fourth scale information and the current candidate data; Performing feature analysis on the historical browsing data in the second scale information to obtain a second interest vector, and performing feature analysis on the feedback behavior in the third scale information to obtain a third interest vector; And obtaining the weight corresponding to each piece of scale information, carrying out weighted fusion on the first interest vector, the second interest vector, the third interest vector and the fourth interest vector according to the weight of each piece of scale information to obtain a fusion vector, and screening the current candidate data according to the fusion vector to obtain screening data as a recommendation result. In a second aspect, an embodiment of the present application provides a data recommendation apparatus, including: The information acquisition module is used for acquiring first scale information, second scale information, third scale information and fourth scale information, wherein the first scale information comprises a label selected by a user, the second scale information comprises historical browsing data, the third scale information comprises feedback behaviors of data in a historical browsing process, and the fourth scale comprises search words of the user in a historical use process; the first interest vector module is used for acquiring current candidate data, determining a first interest vector according to the tag in the first scale information and the current candidate data, and determining a fourth interest vector according to the search word in the fourth scale information and the current candidate data; The second interest vector module is used for carrying out feature analysis on the historical browsing data in the second scale information to obtain a second interest vector, and carrying out feature analysis on the feedback behavior in the third scale information to obtain a third interest vector; The data recommendation module is used for acquiring the weight corresponding to each piece of scale information, carrying out weighted fusion on the first interest vector, the second interest vector, the third interest vector and the fourth i