Session recommendation system and method based on time information and star map network

Abstract

The invention discloses a session recommendation system and method based on time information and a star map network in the technical field of session recommendation, and the method comprises the steps: obtaining a session sequence of an article browsed by a user, and obtaining an initialized source node and a star node; obtaining time feature level information of a source node by using time information of a user browsing an article in the session sequence and a gated graph neural network; calculating the similarity between each source node and each star node by using attention concentration; integrating the information of the source node and the star node based on the similarity of the source node and the star node to obtain a new representation of the source node; updating the star node representation based on the new representation of the source node; a source node and a star node are iteratively updated for multiple times by stacking a multi-layer star map network; using a road network to combine source nodes before and after the multi-layer star map network to obtain a final session representation; and calculating scores of all candidate item items according to the final representation of the session, and generating a recommendation list. The complexity of the network structure is reduced from secondary to linear, and the calculation of the model is greatly reduced.

Description

technical field [0001] The invention relates to a session recommendation system and method based on time information and a star map network, and belongs to the technical field of session recommendation. Background technique [0002] Recommender systems play a key role in various online platforms as they effectively alleviate the problem of information overload by recommending useful content to users. Traditional recommendation methods such as collaborative filtering often rely on the availability of user profiles and long-term historical interactions, and in many recent real-world scenarios, when such information is not available, traditional recommendation methods may perform poorly in terms of predictive performance . Session-based recommendation has attracted extensive attention of researchers in recent years due to its high practical value. It models and predicts the items of interest to the next user according to a given anonymous behavior sequence in chronological ord...

AI Technical Summary

Problems solved by technology

First, almost all session recommendation algorithms based on graph neural networks discard the time information of users browsing items, only sort items according to interaction time, and focus on sequential pattern mining, and the session graphs they construct The edge weights between items are pre-specified and kept fixed before the model is trained, that is, these methods (implicitly) assume that all adjacent items in the sequence have the same time interval, affecting the next item The factors are only the location and

Images