Sentiment Classification from Word of Mouth Documents based on Chinese Collocations

Word of mouth (WOM) has become the main information resource while making business or buying strategies. Most WOM mining research studies focus on classification of WOM documents according to their sentimental orientations, i.e. positive and negative. Generally speaking a well-defined sentiment lexicon is used to provide the sentiment score for words. As a word may have different meanings when used in different domains so it may have different sentiment score. However such a lexicon is static and does not adapt to different domains. In this paper, we first build an adaptive Chinese sentiment lexicon from a real product review website. Then we identify feature words and opinion words of each sentence via the technique of mutual correlations between words. Based on association rules and mutual information, we extract the feature words and their associated collocation words. Finally the term frequency-inverse class frequency (TF-ICF) is used to extract word sentiment scores. According to experimental results, the usage and distribution of words are varied from different domains and our approach has a potential for Chinese WOM classification.