부산대학교 도서관

Recently, user-reviews which have evaluation information for products have been increasing because of the diversity of Internet. Automatical extraction method of beneficial information is useful for both companies and consumers. When we analyze reviews, we usually use the lexicon of sentiment expressions to determine positive or negative sentiment for each word or sentence. There are many researches on generating method of the lexicon of sentiment expressions. The conventional methods use either rating information or local information: local sentiment expressions whose polarities are known. In this study, we combine the rating information and the local information to generate the lexicon of sentiment expressions for improving its accuracy. In this paper, we study how to combine them based on the accuracy of polarities of words.