An empirical study of unsupervised sentiment classification of Chinese reviews

摘要: Abstract This paper is an empirical study of unsupervised sentiment classification Chinese reviews. The focus on exploring the ways to improve performance based limited existing resources in Chinese. On one hand, all available lexicons — individual and combined are evaluated under our proposed framework. other domain dependent noise words identified removed using unlabeled data, performance. To best knowledge, this first such attempt. Experiments have been conducted three open datasets two domains, results show that algorithm for removal can significantly.