Expanding Sentiment Lexicon with Multi-word Terms for Domain-Specific Sentiment Analysis

The increasing interest to extract valuable information from networked data has heightened the need for effective and reliable sentiment analysis techniques. To this end, lexicon-based sentiment classification has been extensively studied by the research community. However, little is known about the usefulness of different multi-word constructs in creating domain-specific sentiment lexicons. Thus, our primary objective in this paper is to evaluate the performance of bigram, typed dependency, and concept as multi-word lexical entries for domain-specific sentiment classification. Pointwise Mutual Information (PMI) was adopted to select the lexical entries and to calculate the sentiment scores of the multi-word terms. With the features generated from the domain lexicons, a series of experiments were carried out using support vector machine (SVM) classifiers. While all the domain-specific classifiers outperformed the baseline classifier, our results showed that lexicons consisting of bigram entries and typed dependency entries improved the performance to a greater extent.