Session recommendation based on graph neural networks incorporating graph attention

Abstract

Session recommendation (RS) is a research focus and hotspot in the field of recommender systems in recent years. Previously, some traditional methods focus on sequence modeling based on user click sequences, but this approach may not be able to fully capture the complex transformation relationships between items. In recent years, graph neural session recommendation introduces session graph topology information to improve the accuracy of item and session feature representation, thus, improving the performance of session recommendation to a certain extent, in addition, this paper also uses graph attention aggregation neighbor nodes to obtain higher-order semantic relationships between items. Based on the item semantic features, the local attention mechanism is used to generate the session features, and finally, based on the item and session semantic features, this paper learns the click probability distribution of different items under a given session by cross-entropy loss. Extensive experiments are conducted on two publicly available and popular datasets, Yoochoose and Diginetica, and the experimental results show that the method has a higher recommendation accuracy.