Thai News Article Sentiment Classification based on User Comments on Online Social Media

Abstract

Social media have become part of our daily life and we read other people's writing including news articles and various user comments and write our own comments. Although most social media services provide a convenient way, LIKE buttons, to express a reactive feeling towards a content, its use does not precisely reflect users' feelings: users often use the button for "LIKE" even when they feel angry, sad, etc. Thus sentiment analysis of users' comments and the recommendation of contents to be presented to each user based on users' comment sentiment analysis is an important issue to improve those services. This thesis proposes a method for this purpose in the Thai language. The Thai language social media text still has difficulties in processing: difficulty in word segmentation with many spelling and other variations included, no publicly available sentiment dictionary, etc. Thus this thesis investigates a systematic noise reduction for social media text using both heuristic rule-based and Conditional Random Fields-based machine learning approaches. Based on this noise reduction method, sentiment dictionary construction based on the data is conducted by comparing different methods. Based on the constructed sentiment dictionary, the sentiment estimation of user comments to news articles is performed. This estimation is employed as labeled sentiment value data for corresponding news articles for each user. Finally news article classification, which is the foundation to recommend future news articles on the basis of each user's preference, is conducted. The final step employs Support Vector Machines as a machine learning method, and sentiment classifiers for each user are constructed. The proposed noise reduction, sentiment dictionary construction, and news article sentiment classification are evaluated with comparative experiments. The result demonstrates that the proposed methods successfully improve each process and the final classification.