Exploiting Sentiment Analysis to Enhance the Collaborative Filtering Recommendations

Abstract

Collaborative Filtering (CF) is considered the most popular and widely employed recommender system approach. However, the CF has exposed limitations such as cold-start and data sparsity. Since the CF approach mostly relies on users' explicit rating preferences to predict items and provided recommendations. Such explicit numerical ratings are usually insufficient or very limited to deliver a good recommendation quality. Recently, many e-commerce websites, such as Amazon, encouraged users to provide comments in free text format, well-known as user reviews to describe their experience with the products. These reviews can also be considered a type of user preference because they usually explain why users liked or disliked a product. This paper proposes to infer the sentiment polarity of user's reviews text using deep learning methods and integrate such preferences of user review sentiment with users actual rating to enhance the recommendation quality of the CF recommender system. The experimental results on different Amazon datasets demonstrate, that the proposed approach improves the performance of the CF recommender system by integrating the sentiment polarity of user reviews in the recommendation process and produces recommendations with higher quality in terms of Recall, precision and F1-measure compared to the baseline CF methods. The results show that the proposed approach achieved state-of-the-art performance, which increased the F1- measure around 3.5%. The Precision by around 3.3% and the Recall around 3.6%, compared with the baseline approaches.