Performance Evaluation of Feature Selection Method for Sentiment Classification of Online Reviews Using Machine Learning Techniques
P. Kalaivani, K. L. Shunmuganathan · International Review on Computers and Software (IRECOS) · 2013
Large volumes of data are available in the web. The discussion forum, review sites, blogs and news corpora are some of the opinion rich resources. More importantly, the information in these forms and reviews are important for both customers and product manufactures, to find the strength and weakness of the product. The information in the Internet is overloaded for customers and unable to read all the reviews and available information. In this study, we evaluate the performance for sentiment classification of online reviews in term of accuracy, precision and recall. We compared three supervised machine learning algorithms of NaivieBayes, Support vector machine and k-NN model for sentiment classification of movie reviews with a size of 2000 documents. The experimental finding indicated that the SVM approach outperformed than the NavieBayes and k-NN approaches, when the number of feature selected about 2000. The SVM approach reached accuracy of more than 83%. We have shown that the number of feature selected between 100 and 1500 for better results.