Examining Negative Sentiment in Reviews to Improve Customer Satisfaction: a New Approach to Recommender Systems

Abstract

Many online product reviews are negative reviews that usually inform other consumers about the quality of the product, but sellers do not benefit much from them. Using natural language processing techniques in this thesis, we will explore how these negative reviews can be used to improve customer satisfaction. Our solution is a newly developed recommender systemthat is able to detect negative sentiment and take this into account for recommending aproduct. We used review data from RateBeer.com and conducted our analysis with feature extraction and classification techniques. User tests revealed that our system scores well on novelty and serendipity, while its trust and diversity scores could use some improvement. Future research ideas include the application of part-of-speech tagging in our approach, or applying our methods to other product categories.