A Rule-Based Chinese Sentiment Mining System with Self-Expanding Dictionary - Taking TripAdvisor as an Example

Abstract

With the wide adoption of social networks, people are accustomed to post their ideas and thinking via these platforms. Tweets or comments online usually come with individual sentiment, which are time consuming to be analyzed by human labor. This study encapsulates a prototype Chinese sentiment mining system and takes a global hotel reviewing website TripAdvisor as the evaluation sample. The proposed sentiment mining model is compared with logistic regression and support vector machine models based on their performances. This proposed model outperforms LR and SVM in all datasets in terms of classification accuracy and F-measure. An additional module embedded in proposed system enables expansion of novel or undefined terms to the dictionary referred (NTUSD). With this Word2Vec-based module, the system further improves accuracy while reduces both type I and type II error for at least 5%.