Convolutional Neural Networks for Sequence-Aware Recommender Systems

Abstract

Recommender systems are prominent components of many of today’s web applications. Historically, the most successful recommender systems have been based on a matrix completion formulation. However, in some domains having sequence-aware recommender systems, i.e systems that take data’s sequential nature into account, may be beneficial to capture user’s short-term interests as well as long-term sequential patterns. The most successful methods for sequence-aware recommender systems have been based on recurrent neural networks. Recurrent neural networks, however, are often hard to train and suffer from several disadvantages in regard to speed and memory requirements. Several recent papers have suggested that convolutional neural networks can be used to process sequential data more efficiently and sometimes with better results than recurrent networks. In this thesis, we propose the use of convolutional neural networks for the task of sequence-aware recommendations. We present a two-stage deep learning approach to recommendations, where convolutional neural networks are used for sequence-aware candidate generation. Our results show that convolutional neural networks can achieve predictive performance comparable to state-of-the-art for sequence-aware recommendation tasks.