Session-based Graph Convolutional ARMA Filter Recommendation Model

With the rapid popularity of Internet shopping, session-based personalized recommendations have become an important means to help people discover their potentially interesting items in real-time. Most existing works only model a session as a sequence and use recurrent neural networks for recommendation. Despite their effectiveness, the results may not be sufficient to capture the potential relationships between items. In this work, we integrate a graph convolutional layer based on Auto-Regressive Moving Average (ARMA) filters into the Graph Neural Network (GNN). In particular, it is a new session-based recommendation framework Graph Convolution ARMA Filter (AUTOMATE), which can capture complex transformations between items through a sequence of sessions modeled as graph-structured data. Each session is then represented as the composition of the current interest and the global preference of that session using an attention network. The rationality and efficacy of the proposed AUTOMATE model are extensively evaluated on two public real-world datasets. Experimental results show that our model is significantly better than other state-of-the-art methods.