Learning-Based Method with Valence Shifters for Sentiment Analysis

Automatic sentiment classification is becoming a popular and effective way to help online users or companies process and make sense of customer reviews. In this article, a learning-based method for classification of online reviews that achieves better classification accuracy is obtained by (a) combining valence shifters and opinion words into bigrams for use as features in an ordinal margin classifier and (b) using relational information between unigrams/bigrams in the form of a graph to constrain the parameters of the classifier. By using these two components, it is possible to extract more information present in the unstructured data than other methods such as support vector machines and random forest, hence gaining the potential of better classification performance. Indeed, our simulation results show a higher classification accuracy on empirical real data with ground truth and on simulated data.