Combining Collaborative Filtering and Sentiment Classification for Improved Movie Recommendations

Abstract

Recommender systems are traditionally of following three types: content-based, collaborative filtering and hybrid systems. Content-based methods are limited in their applicability to textual items only, whereas collaborative filtering due to its accuracy and its black box approach has been used widely for different kinds of item recommendations. Hybrid method, the third approach, tries to combine content and collaborative approaches to improve the recommendation results. In this paper, we present an alternative approach to a hybrid recommender system that improves the results of collaborative filtering by incorporating a sentiment classifier in the recommendation process. We have explored this idea through our experimental work in movie review domain, with collaborative filtering doing first level filtering and the sentiment classifier performing the second level of filtering. The final recommendation list is a more accurate and focused set.