Deep Learning-Based Recommender Systems

Abstract

The term “information overload” has gained popularity over the last few years. It defines the difficulties people face in finding what they want from a huge volume of available information. Recommender systems have been recognized to be an effective solution to such issues, such that suggestions are made based on users’ preferences. This chapter introduces an application of deep learning techniques in the domain of recommender systems. Generally, collaborative filtering approaches, and Matrix Factorization (MF) techniques in particular, are widely known for their convincing performance in recommender systems. We introduce a Collaborative Attentive Autoencoder (CATA) that improves the matrix factorization performance by leveraging an item’s contextual data. Specifically, CATA learns the proper features from scientific articles through the attention mechanism that can capture the most pertinent parts of information in order to make better recommendations. The learned features are then incorporated into the learning process of MF. Comprehensive experiments on three real-world datasets have shown our method performs better than other state-of-the-art methods according to various evaluation metrics. The source code of our model is available at: https://github.com/jianlin-cheng/CATA.

This chapter is an extended version of our published paper at the IEEE ICMLA conference 2019 [1]. This chapter incorporates new experimental contributions compared to the original conference paper.