A recommender system with sentiment classification of product reviews

Abstract

Product reviews are an important factor when considering a purchasing decision. Thus, people carefully read and analyze them. However, if the amount of reviews grows too large, customers have to spend a long time in selecting the goods they wish to buy. To solve this problem, in this paper we propose a recommender system with sentiment classification for product reviews. Using sentiment classification and a number of product reviews, we implement a recommender system that considers the two most widely used recommendation methods, collaborative and content-based filtering, by quantizing the sentiments included in the reviews. Also, to improve the performance of the proposed system, we analyze the product reviews to extract sentiment-related terms a semantic orientation, and suggest a heuristic weighting scheme. In our first experiment, we verify the weighting scheme and the extracted terms. As a result, the proposed sentiment classification system shows about 90% accuracy. A second experiment is for verifying the recommender system. The recommender system ranks products using the scores obtained from the weighting scheme and the number of reviews. In the experiment, the proposed system shows better results than recommendations based on consumer ratings.