Update method and device, storage medium and computing device of behavior prediction system

Abstract

The embodiment of this specification provides a method for updating a behavior prediction system, the method includes: obtaining training sample pairs, including source domain samples and target domain samples corresponding to the same user, each domain sample includes the current object of the corresponding domain, and the user has done M historical objects with specific behaviors, and a behavior label indicating whether the user has made a specific behavior on the current object; the training samples are input to the behavior prediction system, and the source domain encoding layer and the target domain encoding layer adopt joint attention Force mechanism, coded to obtain the current object vector in the source domain and M historical object vectors in the source domain, as well as the current object vector in the target domain and M historical object vectors in the target domain; based on 2M historical object vectors, perform the first fusion and the second fusion respectively , to obtain the source domain user vector and the target domain user vector correspondingly; based on the user vector of each domain and the current object vector of each domain, determine the behavior prediction results of each domain, and then combine the behavior labels of each domain to update the behavior prediction system.

Description

technical field [0001] The embodiments of this specification relate to the field of computer technology, and in particular to a method and device for updating a behavior prediction system. Background technique [0002] Nowadays, people are using various services provided by service platforms to users more and more frequently. Correspondingly, in order to improve the service experience of users, service platforms can use machine learning models to predict relevant behaviors of users when using services. Then, according to the prediction results, the service plan is customized for the user. For example, a news information website may determine the category and order of the news sections included in the news information page pushed to the user by predicting the click probability of a user on various news sections. For another example, a shopping website can determine whether to recommend a product to a user by predicting a user's liking for a product. [0003] Obviously, it i...

AI Technical Summary

Problems solved by technology

However, the current method of predicting user behavior is relatively single, and the accuracy of the prediction results obtained is also very limited.

Images