Social Recommendation Using Deep Auto-encoder and Confidence Aware Sentiment Analysis

The development of online social networks has attracted increasing interest in social recommendation. On the other hand, recommender systems based on deep learning and sentiment analysis techniques are currently widely used to solve the problem of data sparsity. However, only a few attempts have been made in social-based recommender systems. This article focuses on this issue and proposes a novel hybrid approach named CASA-SR (Confidence Aware Sentiment Analysis-based Deep Social Recommendation). Our approach exploits sentiment analysis by detecting fake reviews and combines predictions generated by collaborative and content-based filtering. A neural architecture has been adopted using an auto-encoder and a multilayer perceptron neural network. Moreover, our approach integrates social information, including users’ trust (credibility and similarity degrees). Experimental results conducted on different datasets showed significant improvements in recommendation performance according to the state-of-the-art work.