Global Context Enhanced Item Learning for Session-based Recommendation

Abstract

Session-based recommendation is a challenging task, which aims at recommending items based on user behavior sequences. Robust item representation learning has a great impact on improving recommendation performance. Existing session-based recommendation methods randomly initialize items into low-dimensional vectors, and then simultaneously train the item representations when training the session-based recommendation model. Existing approaches do not make full use of user historical interaction data. In order to make full use of user interaction data, we first construct a global item transition graph based on all user interaction data, and then apply a graph neural network to learn the item representation in the item transition graph. The downstream session-based recommendation models learn user preference and give recommendation results based on the global item representation. Our proposed global context item learning method can effectively utilize the global item transition knowledge, thus can learn rich item representations. We conduct inclusive experiments on three common datasets, and the results show the effectiveness and the efficiency of our proposed framework.