Session recommendation method and system based on structure and semantic attention stacking

Abstract

The invention discloses a session recommendation method and system based on structure and semantic attention stacking, and belongs to the technical field of session recommendation. According to the method, on the basis of a user article click sequence, firstly, an attention mechanism model based on semantics, a time sequence structure and a space structure is established, the importance degree of related articles on user global interest characterization is enhanced from different angles, and interference of irrelevant articles on user global interest characterization is weakened; and user global interests under three angles are established; then, three user global interest fusion models are established, and user global interest characterization at multiple angles is formed; and finally, in combination with user local interest characterization generated based on the GRU recurrent neural network, user interest characterization is established, and prediction of articles clicked by the user at the next moment is realized.

Description

technical field [0001] The invention relates to the technical field of conversation recommendation, in particular to a conversation recommendation method and system based on structural and semantic attention stacking. Background technique [0002] The recommendation system realizes invalid content filtering and useful content recommendation for users under the background of information overload. This function has been applied to many online platforms and plays an important role in promoting user consumption and increasing sales, such as Taobao, JD.com Wait. Traditional recommendation methods are mostly implemented based on user information and their long-term historical behavior, but they cannot be effectively satisfied in practical application scenarios, such as lack of user information due to lack of user registration and login. To this end, the academic community has proposed a conversational recommendation task, that is, by mining the behavior information of anonymous u...

AI Technical Summary

Problems solved by technology

However, when modeling the user's global interest, this method only highlights the semantic association information of the last item compared with the item sequence, without considering the importance of temporal structure, spatial structure and other information to the user's global interest representation, which affects Accuracy and Rationality of User Interest Representation

For this reason, this patent proposes a conversation recommendation method and system based on structure and semantic stacking to solve the problem of inaccurate user interest representation

Images