Lexicon-based Chinese opinion analysis on online product reviews

Online product reviews contain opinions and sentiments from the Internet users. These opinions are useful for brand analysis or public opinion analysis. In this paper we propose a lexicon-based opinion analysis method on Chinese product reviews. The lexicon is built automatically from the corpus, where feature words, out-of-vocabulary words, and feature-opinion pairs are extracted. Then based on this lexicon, text-based features are extracted for a supervised learning of a multi-stage sentiment classification. The tasks include a sentence-level opinion analysis and a document-level opinion rating task. The proposed method achieves an accuracy of approximately 50% on the three-class classification in the first task, and an accuracy of 40%-60% on the five-class classification in the second task. © 2013 by Binary Information Press.