An Ontology-Based Approach to Sentiment Classification of Mixed Opinions in Online Restaurant Reviews

Consumers review other consumer's opinion and experience of the quality of various products before making purchase. Automatic sentiment analysis of WOM in the form of user product reviews, blog posts and comments in online forum can support strategies in areas such as search engines, recommender systems, and market research and benefit to both consumers and sellers. The ontology-based approach designed in this work aims to investigate how to detect and classify mixed positive and negative opinions by interpreting with an ontology containing opinion information on terms. Our research question is whether disinterested subjectivity scores of sentiment ontology are pertinent to sentiment orientations not affected by reviewer's linguistic bias. The experimental results adopting opinion lexical resource achieve better and more stable performance in F-measure.