Improve Sentiment Analysis Accuracy using Multiple Kernel Approach
Rinki Sharma, Amit Arora · International Journal of Computer Applications · 2013
Sentiment Analysis has become an indispensible part of product reviews in present scenario. Sentiment Analysis is a very well studied field, but the scale remains limited to not more than a few hundred researchers. The problem of analyzing the overall sentiment of a document using Machine learning techniques has been considered. Results have been improved using multiple kernel approach and compared with previously used techniques..The present research is a comparison and extension of the work proposed by Mullen and Collier (2003). The system consists of a feature Extraction phase and a learning phase; on the basis of which the overall sentiment of the document is analyzed. The present work uses the movie review data set used by Pang (2002). The approach significantly outperforms the previous methods attaining 90% and 92% accuracy using 5 fold cross validation 10 fold cross validation respectively. General Terms Machine Learning, Information Retrieval.