The naive Bayes classifier in opinion mining: In search of the best feature set

Abstract

This paper focuses on how naive Bayes classifiers work in opinion mining applications. The first question asked is what are the feature sets to choose when training such a classifier in order to obtain the best results in the classification of objects (in this case, texts). The second question is whether combining the results of Naive Bayes classifiers trained on different feature sets has a positive effect on the final results. Two data bases consisting of negative and positive movie reviews were used when training and testing the classifiers for testing purposes. © 2012 Springer-Verlag.