Information recommendation method, device, server and storage medium

Abstract

The embodiment of the present invention discloses an information recommendation method, device, server, and storage medium, wherein the method includes acquiring the reference user vector of the target user in the reference field, and calling the mapping model to perform calculations on the reference user vector to obtain the target user The target user vector in the target field, according to the target user vector to obtain recommendation information for the target user, and recommend the recommendation information in the target field. Among them, the mapping model is obtained by training the basic mapping model through the reference sample vector set of the sample user set in the reference field and the target sample vector set in the target field, and the basic mapping model is the initial mapping model trained by K sets of training vector sets It is obtained that different types of data are used to complete the training of the mapping model at different stages, and the information recommended to users is determined based on the mapping model, which improves the accuracy of information recommendation.

Description

technical field [0001] The present invention relates to the field of computer technology, in particular to an information recommendation method, device, server and storage medium. Background technique [0002] With the development of Internet technology, users can obtain a variety of information through different fields (applications, public accounts, video accounts, etc.). Record and analyze, get the information that the user is interested in (text, picture, video, etc.), and recommend the information that the user is interested in to the user in the field. However, for a new user in a field, if there is no interaction data between the user and the field, the browsing records of the user in this field cannot be obtained, so that it is impossible to cold start the user based on the user's records in this field. recommend. [0003] At present, each field has its own recommendation system. In the recommendation system, each user is represented by a vector, and the vector rep...

AI Technical Summary

Problems solved by technology

In order to solve the problem of cold start information recommendation, a mapping model is currently implemented to learn the mapping of user vectors in different recommendation systems, so that the vector representation of users in one recommendation system can be transferred to another recommendation system. For example, when a user just uses the target In the domain, the vector representation of the user in the reference domain is obtained, and the vector representation in the reference domain is processed through the mapping model to obtain the vector representation of the user in the target domain. The model training in the above method requires a large number of Both the reference domain and the target domain have interactive sample user data, and the model is trained based on the vector representations of the sample users in different domains. If there is interaction with the target field, the mapping model trained based on a small number of sample users will have a low mapping accuracy in the actual application process, which will lead to a mismatch between the information recommended by the user vector based on the mapping and the user's interests.

Images