Feature-Wise Opinion Summarization of Consumer Reviews Using Domain Ontology

Abstract

There is a rapid increase of contents generated by users such as reviews and comments over the Internet. Analyzing a review is critical for data-driven decision making and corporate intelligence of individuals and organizations. This paper focuses on feature-wise sentiment analysis of reviews which results in a summary based on the important features of a specific domain. The proposed approach consists of two major processes: data acquisition and preprocessing to aggregate reviews from different sources. Ontology is generated based on the domain-specific knowledge, and it is updated with new features, opinions and sentiment orientation. Sentiment determination process uses the lexicon, SentiWordNet to discovery the sentiment orientation of each opinion word and PMI value to find the sentiment expectation of each new feature. The summarization of reviews represents the feature-level summary of reviews to the user. Evaluation results show high precision and recall values which measures the exactness and completeness of the approach.