Dynamic Feature Subsumption based Multiclass Sentiment Analyzer using Machine Learning Techniques

Abstract: Sentiment classification on product reviews which has been studied earlier concluded that the subjective text consists of either positive and negative opinions but the user reviews can be classified in to broader level(multiclass) which gives deeper view of a user in more than two classes .Multiclass classification has been done for different domains using the overall ratings for a product given by the user, but nobody concentrated on classifying each user opinion in to multiple categories using the scoring function. We proposed a novel methodology to get multiclass sentiment labels for each textual comment considering each feature of the product that too for each sentence in the comment. Maxentropy parts of speech tagger proposed by Stanford University is used in our work to extract tagged features from the text which are used in identifying opinion of a user. Our work improved the original porter stemmer algorithm by adding new rules so as to improve overall performance. Results are evaluated on training and testing data when given to machine learning algorithms, our approach got high percentage of accuracy when is compared with other existing works.

Keywords: Sentiment classification, machine learning techniques, feature subsumption, Natural language processing(NLP), sentiwordnet, support vector machine(svm).