Effective Use of Linguistic Features for Sentiment Analysis of Korean

Effective Use of Linguistic Features for Sentiment Analysis of Korean* Hayeon Jang a and Hyopil Shin b Department of Linguistics, Seoul National University, 599 Gwanak-ro, Gwanak-gu, Seoul, Republic of Korea ahyan05@snu.ac.kr bhpshin@snu.ac.kr Abstract. In this paper, we propose a new linguistic approach for sentiment analysis of Korean. In order to overcome shortcomings of previous works confined to statistical methods, we make effective use of various linguistic features reflecting the nature of Korean such as contextual intensifiers, contextual shifters, modal affixes, and the morphological dependency chunk structures. Moreover, unlike complex statistical formulae which are hard to understand, we use simple mathematical formulae in the process of term weighting. Through experiments of news corpus, we verify an improvement on the results of sentiment analysis of Korean in comparison to the experimental results using TFIDF as popular statistical method employing word frequency. This approach, especially the chunking method, will be beneficial to sentiment analysis of other morphologically rich languages like Japanese and Turkish. Keywords: Sentiment analysis, Linguistic feature, Morphologically rich language * Copyright 2010 by Hayeon Jang and Hyopil Shin 1 Introduction The Internet is now an important forum where people can express their opinions without formality and constraint. Online services such as blogs and Twitter replace private diaries, and political debates are frequently opened in the reply pages of news articles. For this reason, sentiment analysis which automatically extracts subjectivities and classifies sentiments (or polarities) about some topics in written texts has been receiving attention in the field of NLP. Moreover the techniques of sentiment analysis are applied to various applications for extracting important information on the Internet to monitor a certain brand’s reputations or to make social network for peoples who have similar opinion. Sentiment analysis of English employs various statistical and linguistic methods. In the case of Korean, however, most previous research has been confined to statistical methods which only focus either on the frequency of words or relevance of co-occurring words, due to the lack of linguistic resources properly reflecting the nature of Korean. The major drawback of a statistical-based approach is the fact that the ‘real’ meaning of expressions, which we feel when we read them, cannot be reflected in the analysis. Moreover, complex mathematical formulae used in statistical methods are difficult for non-specialists in statistics or mathematics to understand and, in turn, require a heavy workload. In order to overcome such shortcomings of statistical methods, we propose a new linguistic approach for sentiment analysis of Korean. The focus of our approach is an effective utilization of various linguistic features of Korean such as contextual intensifiers, contextual shifters, modal affixes, and chunk structures. To explain briefly: contextual intensifiers strengthen the valence of opinionated terms, contextual shifters change sentimental flow of sentences or polarity of terms, modal affixes determine whether or not situation included in the sentence is PACLIC 24 Proceedings 173 true, and chunk structures limit influential scopes of negation items. Through experiments using news corpus, we verify an improvement on the results of sentiment analysis. One of …

Content truncated to 500 words