Objective Words Can Improve Sentiment Classification for Word of Mouth

Word of mouth (WOM) has a strong effect on consumer behavior. A sentimental lexicon, i.e. SentiWordNet is useful for the task of WOM sentiment classification. However, most current related research ignores the problem that too many objective words are defined in SentiWordNet. In this research, we focus on the effect of objective words on the performance of WOM sentiment classification, and propose a novel sentimental relevance approach. We analyze the co-relation of each objective word and its associated sentences in order to modify the sentimental score and tendency for the objective word. This semantic-oriented approach integrated with a machine learning-oriented approach, i.e. support vector machine, is capable of making an improvement in WOM sentiment classification.