Context-Assisted Sentiment Analysis

This paper explores how information regarding the context can assist in improving sentiment analysis performance. We postulate that context affects sentiment at two different levels: first at the domain level of the comment, and second, at the sentence-structure level. Noting this, we explore three ways of utilizing context in sentiment analysis. First, we study contextual assistance with respect to the domain through ontology support for a class of sentiment analysis approaches, namely, sentiment dictionary based methods, which are primarily domain-context-free. Here, an architecture is proposed to build a domain ontology, by automatically mapping comment sentences to objects in the ontology and construct sentiment analysis based on the ontology. Second, we introduce a new method of utilizing sentence structure context in sentiment analysis and use this method to assist machine-learning approaches. Third, we also propose two architectures to combine multiple sentiment methods where the rationale for the combination is context-driven. Finally, experiments show that our solutions improve performance over baseline approaches.