Instance-Based Enrichment of Sentiment Ontology

Abstract

Sentiment ontology is usually realized as triplets of objects, aspects, and sentiment of a product. Such ontology can effectively support coreference resolution in textual documents. To develop a sentiment ontology, one can start from a preliminary comprehensive knowledge base and gradually enrich it. This paper proposes a method to a sentiment ontology, which is based on the instance relationship captured from collected data. Our approach is a hybrid one based on rules, dependence grammar, and existing knowledge bases, such as WordNet and Opinion Lexicon. The application of a dependency grammar is to find relationships between components in the ontology and WordNet is used to identify semantic classes for unknown words and phrases. We have experimented our approach with real datasets and the preliminary results are quite promising.