Sentiment Classification for Movie Reviews in Chinese Using Parsing-based Methods

International Joint Conference on Natural Language Processing, pages 561–569, Nagoya, Japan, 14-18 October 2013. Sentiment Classification for Movie Reviews in Chinese Using Parsing-based Methods Wen-Juan Hou and Chuang-Ping Chang Department of Computer Science and Information Engineering, National Taiwan Normal University, No.88, Sec. 4, Tingzhou Road, Wenshan Distinct, Taipei, Taiwan emilyhou@csie.ntnu.edu.tw; lukechang@std.ntnu.edu.tw Abstract Sentiment classification is able to help people automatically analyze customers’ opinions from the large corpus. In this paper, we collect some Chinese movie reviews from Bulletin Board System and aim at making sentiment classification so as to extract several frequent opinion words in some movie elements such as plots, actors/actresses, special effects, and so on. Moreover, we result in a general rec- ommendation grade for users. Focusing on the movie reviews in Chinese, we propose a novel procedure which can extract the pairs of opin- ion words and feature words according to de- pendency grammar graphs. This parsing-based approach is more suitable for review articles with plenty of words. The grading results will be presented by a 5-grade scoring system. The experimental results show that the accuracy of our system, with the deviation of grades less than 1, is 70.72%, and the Mean Reciprocal Rank (MRR) value is 0.61. When we change the 5-grade scoring system into producing two values: one for recommendation and the other for non-recommendation, we get precision rates 71.23% and 55.88%, respectively. The result shows an exhilarating performance and indicates that our system can reach satisfied expectancy for movie recommendation. 1 Introduction As the rapid growth of text data, text mining has been applied to discover hidden knowledge from text in many applications and domains. Nowa- days, reviews are increasing with a rapid speed and are available over internet in natural lan- guages. Sentiment analysis tries to identify and extract subject information from reviews. The problem of automatic sentiment analysis has re- ceived significant attention in recent years, large- ly due to the explosion of online social-oriented content (e.g., user reviews, blogs, etc). Further- more, sentiment analysis can be used in various ways and in many applications such as sugges- tion systems based on the user likes and ratings, recommendation systems, or insisting in election campaigns. As one of the important applications, sentiment classification targets to rate the polari- ty of a given text accurately towards a label or a score, predicting whether the expressive opinion in the text is positive, negative, or neutral. Identifying the sentiment polarity is a complex task. To address the problem of sentiment classi- fication, various methodologies have been ap- plied earlier. Generally, there are two types of approaches tackling the sentiment classification task according to the knowledge the systems used. One is corpus-based and the other is lexi- con-based. Corpus-based approaches are usually supervised, i.e., requiring training sets, and per- forming well when the training set is large enough and correctly labeled. The approaches are studied in (Bakliwal et al., 2011; Bespalv et al., 2011; Kennedy and Inkpen, 2006; Jin et al., 2009; Narayanan et al., 2009; Pang …

Content truncated to 500 words