Cross-domain sequence recommendation method based on adaptive fine tuning strategy

Abstract

The invention provides a cross-domain sequence recommendation method based on an adaptive fine tuning strategy. The method comprises the steps of constructing a sequence recommendation model, trainingthe sequence recommendation model according to behaviors of a user in a source field, and obtaining a pre-trained sequence recommendation model used for reflecting behavior habits of the user in thesource field; adjusting a part of trained parameters of the pre-trained sequence recommendation model in combination with the behavioral habits of the user in the target field to obtain an adjusted sequence recommendation model; and in the target field, according to a given user behavior sequence, recommending to the user by utilizing the adjusted sequence recommendation model. According to the method, cross-domain sequence recommendation can be realized, and the robustness and accuracy of cross-domain sequence recommendation are improved.

Description

technical field [0001] The invention relates to the technical field of sequence recommendation, in particular to a cross-domain sequence recommendation method based on an adaptive fine-tuning strategy. Background technique [0002] With the popularity of Internet services such as e-commerce, online service platforms, and online transactions, the analysis and recommendation of online user behavior has become a hot research topic. For example, a session is a mechanism used by the server to identify a user. If a user clicks on a series of products or browses a series of web pages, the server creates a specific session for the user and tracks their click and browse behavior. It can be understood that a session is A sequence of user browsing records with a temporal relationship. Sequence recommendation system (or session recommendation system) is an important branch of recommendation system, its purpose is to make accurate recommendations to users by analyzing the user's histori...

AI Technical Summary

Problems solved by technology

[0004] The existing technology mainly has the following problems: 1). In the process of pre-training the sequence recommendation model, it is impossible to pre-train a sequence recommendation system with good performance on the source field with rich browsing records of users, and it cannot model users well. Browsing habits in the source field; 2) Selecting and fine-tuning all the parameters in the process of fine-tuning the pre-training model is a very time-consuming and storage-intensive work. The model parameters are huge, and the training is very time-consuming and consumes a lot of storage space. Meet the actual requirements; 3), the model that fine-tunes all parameters in the fine-tuning process is easy to overfit during the training process, the training is very unstable, and it is easy to lose important information in the pre-training model, and the robustness is poor

Images