Polarity Identification of Sentiment Words Based on Emoticons

The orientation of sentiment words plays an important role in the sentiment analysis, but existing methods have difficulty in classifying the orientation of Chinese words, especially for the newly emerged words in Internet. Most approaches are mining the association between sentiment words and seed words using the big corpora and manually labeled seed words with definite orientation. But less work has ever focused on the efficient seed words selection. As we observed, emoticons, which are widely used on social network because of the simplicity and visualization, are good indicators for sentiment orientation. Thus this paper proposes the sentiment word model based on emoticons, which built orientation model of sentiment words with the orientation of emoticons, and train the model with the SVM classifier. Meanwhile, this work proposes a high efficient way to automatically classify the orientation of emoticons. Experiments show the precision rate of emoticon classification could reach 93.6%, and that of sentiment words classification could be 81.5%.