The application of semantic domain representation to the problems of sentiment analysis and opinion mining

Date

Authors

Coe, Benjamin

Journal Title

Journal ISSN

Volume Title

Publisher

University of Guelph

Abstract

In this thesis an approach to Sentiment Analysis and Opinion Mining is put forward that combines three computational tasks: the semantic representation of a domain (using the semantic markup language, OWL), the identification of entities within a domain, and the application of a sentiment analysis algorithm. The approach is applied to classifying film reviews based on a "thumbs-up" or "thumbs-down" rating. The approach was applied to a human-tagged corpus of 174 film reviews, and to a computer-tagged corpus of 1087 film reviews, and achieved 94% and 74% accuracy respectively. The experiment was undertaken in the hopes of proving the hypothesis that domain specific representation could vastly improve an existing Sentiment Analysis and Opinion Mining approach. The experiment was seen as being a success, improving upon the baseline approach outlined in [29] by 28% with the human-tagged corpus, and 11% with the computer-generated corpus.

Description

Keywords

Sentiment Analysis and Opinion Mining, computational tasks, semantic representation, domain, entities, sentiment analysis

Citation