Time and Position Aware Graph Neural Networks for Session-based Recommendation

Abstract

Session-based recommendation system is an important part of many e-commerce sites. Its purpose is to recommend according to the interaction behavior of anonymous users in a short time. The latest research is to model the session sequence as a graph and then use the graph neural network to learn the embedding of the item. However, these methods treat the session as a simple sequence and ignore the time interval between user’s adjacent interactions. In order to solve this problem, we propose a Time and Position Aware Graph Neural Networks model for the session-based recommendation systems, which can not only learn the embedding of items, but also capture users' interest by using the time interval and sequence information when users browse items. We have conducted sufficient experiments on two e-business datasets, and the experimental results show that our model is superior to baselines.