Handling polarity shift for opinion mining

Opinion mining or sentiment analysis is the study of people's opinion, sentiments, attitudes and emotions expressed in written language. It is one of the most active research programs in natural language processing and text mining. Though sentiment analysis has been a trending research program in the community of natural language processing, the Bag Of Words based machine learning approach is state-of-the-art for this task. However, BOW model does not focus much on polarity shift which may create a different overall impact. Polarity shift handling is one of the major problems in performing sentimental analysis of any text or sentence. Earlier work has been done on handling the polarity shifts focused on detecting polarity shift with limited scope. Some works also included training of the classifier either by reversing of the original reviews or extracting the features based on patterns. This study aims to handle polarity shift. Sentiment classification will be performed in 2 major steps i.e. Tokenization and Polarity shift handling. A model is proposed to handle explicit negation with a larger scope. Unlike traditional negation modifiers, our aim is to negate the related terms even if it does not immediately follow negation modifier. Apart from modification and polarity shift handling, other tasks for sentimental classification of tokenization are performed in a traditional way. Proposed model is evaluated on Kaggle's “Bag of Words meets Bag of Popcorn” which is a balanced dataset consisting of 25000 positive and negative reviews in total. For the proposed model, partial data-set have been used. 10-fold cross validation technique is used for evaluation of proposed model. Proposed model is compared and analysed with the existing state-of-the-art model such as PSDEE.