Detecting Domain Dedicated Polar Words

International Joint Conference on Natural Language Processing, pages 661–666, Nagoya, Japan, 14-18 October 2013. Detecting Domain Dedicated Polar Words Raksha Sharma, Pushpak Bhattacharyya Dept. of Computer Science and Engineering IIT Bombay, Mumbai, India {raksha,pb}@cse.iitb.ac.in Abstract There are many examples in which a word changes its polarity from domain to do- main. For example, unpredictable is pos- itive in the movie domain, but negative in the product domain. Such words cannot be entered in a “universal sentiment lexi- con” which is supposed to be a repository of words with polarity invariant across do- mains. Rather, we need to maintain sep- arate domain specific sentiment lexicons. The main contribution of this paper is to present an effective method of generat- ing a domain specific sentiment lexicon. For a word whose domain specific polar- ity needs to be determined, the approach uses the Chi-Square test to detect if the dif- ference is significant between the counts of the word in positive and negative po- larity documents. We extract 274 words that are polar in the movie domain, but are not present in the universal sentiment lex- icon. Our overall accuracy is around 60% in detecting movie domain specific polar words. 1 Introduction Sentiment analysis (SA) has attracted a great deal of attention in recent times (Hatzivassiloglou and McKeown, 1997; Wiebe, 2000; Pang et al., 2002; Turney, 2002; Yu and Hatzivassiloglou, 2003; Hu and Liu, 2004; Esuli and Sebastiani, 2005; Breck et al., 2007). The SA task is to pre- dict the sentiment orientation of a text (docu- ment/para/sentence) by analyzing the polarity of words present in the text. A lexicon of sentiment bearing words is of great help in such tasks. Sentiment lexicons are of two types: univer- sal and domain specific. Words like ‘good’ and ‘bad’ have uniform polarity across all domains, and so are members of universal sentiment lex- icon. A word like ‘unpredictable’, on the other hand, is positive in the movie domain (‘unpre- dictable plot’), but negative in the car domain (‘un- predictable steering’). Such a word should be en- tered as positive in the sentiment lexicon of the movie domain and as negative in the sentiment lexicon of the car domain. There are many “universal sentiment lexi- cons” like SentiWordNet1, subjectivity lexicon2 by Wiebe, list of positive and negative opinion words3 by Liu. These lexica contain only those polar words which have the same polarity in all domains. In this paper, we use the universal senti- ment lexicon published by Wiebe. Using resources like Wikipedia and SentiWord- Net to determine polarity of a domain specific word may lead to wrong sentiment detection. The motivation for our work comes from addressing this problem. We would like to create domain spe- cific sentiment lexicons. Our technique for detecting domain specific po- lar words is inspired by the work done by Cheng and Zhulyn (2012). They used the Pearson’s Chi- Square test to find the top 200 words most indica- tive of positive sentiment and the top 200 words most indicative …

Content truncated to 500 words