The Development of Semi-automatic Sentiment Lexicon Construction Tool for Thai Sentiment Analysis

摘要

Sentiment analysis has gained so much interest from many companies and organizations in Thailand. However, there are a few research studies focused on developing Thai sentiment lexicon, which is an important resource for the sentiment analysis. In this work, we developed a web-based automatic Thai lexicon construction tool. Our tool employed a semi-supervised approach for semi-automatically extracting the sentiment lexicon entries. To reduce a negative impact from unreliable parser, we provide simple heuristic rules and mutual information for recognizing sentiment words and its features. The polarity of recognized sentiment words is automatically identified through a bootstrapping process that utilizes a small set of sentiment seeds, the context coherency characteristics, and statistical co-occurrence. In the evaluation, we received quite fair results for lexicon construction task, 76.06 and 75.28 F-Score for hotel review and laptop review, respectively.