The study of sentiment analysis on Chinese comment texts

摘要

As known to all, a word's implication could differ in different domain, even the same word in the same domain may indicate diverse polarity with respect to different aspects. Due to the domain-dependent characteristic of some words, it is difficult to build a generally used sentiment lexicon. At the same time, there are a large amount of terms whose polarity is always definite. For example, “excellent” and “bad” respectively represent determinate polarity of “positive” and “negative”. In this paper, we construct a sentiment lexicon making use of both the context-dependent and the context-independent features. The traits of the semi-structured product reviews offered by users are also used in the construction process. At last, we put forward a method of combining the domain-specific lexicon and the existing sentiment wordlist such as “Hownet” for sentiment analysis.