Generating Domain-Specific Ontology from Common-Sense Semantic Network for Target-Specific Sentiment Analysis

Target or feature specific sentiment classification of a product review consists of extracting opinion or sentiment expressing phrases, extracting the targets (features in a product domain), computing the semantic orientation of the sentiment expressing phrase and assigning the sentiment expression to the product feature it targets. Each of the tasks is fundamental to the problem of target-specific sentiment analysis. In this paper, we present an algorithm to automatically build a domain-specific ontology (a graph consisting of product features and semantic relations between them) which can be used as a lexical resource for performing target-specific sentiment analysis in real-time. We use ConceptNet (a large semantic network of commonsense knowledge) for extracting domain-specific ontology. We evaluate our approach on publicly available preannotated dataset from phone and camera domain. The advantages of our approach are that it uses a resource which is created by volunteers on the Internet and not by trained or specialized knowledge engineers. Another advantage is the product feature lexicon that is created is in the form of semantically rich domain ontology rather than a flat list of phrases. We investigate the usefulness of commonsense knowledge for generating domainspecific ontology for feature extraction task in sentiment analysis application and conclude that the ap-