Aspect-Based Sentiment Analysis Using Lexico-Semantic Patterns

With its ever growing amount of user-generated content, the Web has become a trove of consumer information. The free text format in which most of this content is written, however, prevents straightforward analysis. Instead, natural language processing techniques are required to quantify the textual information embedded within text. This research focuses on extracting the sentiment that can be found in consumer reviews. In particular, we focus on finding the sentiment associated with the various aspects of the product or service a consumer writes about. Using a standard Support Vector Machine for classification, we propose six different types of patterns: lexical, syntactical, synset, sentiment, hybrid, surface. We demonstrate that several of these lexico-syntactic patterns can be used to improve sentiment classification for aspects.