CATA++: A Collaborative Dual Attentive Autoencoder Method for Recommending Scientific Articles

Abstract

Matrix Factorization (MF) method is widely popular for personalized recommendations. However, the natural data sparsity problem limits its performance, where users generally only interact with a small fraction of available items. Accordingly, several hybrid models have been proposed recently to optimize MF performance by incorporating additional contextual information in the MF's learning process. Although these models improved recommendation quality, there are two primary aspects for further enhancements: (1) multiple models focus only on some portion of the available contextual information and neglect other portions, and (2) learning the feature space of the contextual information needs to be further improved. Therefore, we introduce a Collaborative Dual Attentive Autoencoder (CATA++) method that utilizes an item's content and learns its latent space via two parallel autoencoders. We employ the attention mechanism in the middle of our autoencoders to capture the most significant segments of contextual information, which leads to a better representation of the items in the latent space. Extensive experiments on three scientific-article datasets have shown that our dual-process learning strategy has significantly improved MF performance in comparison with other state-of-the-art MF-based models using various experimental evaluations. The source code of our method is available at: https://github.com/jianlin-cheng/CATA.

Keywords

• Autoencoder
• collaborative filtering
• deep learning
• matrix factorization
• recommender systems