Most Persistent Feature Selection Method for Opinion Mining of Social Media Reviews

Abstract

Many business organizations use social media data in order to understand their customer on an individual level. Consumers are keen to share their views on certain products or commodities. This leads to the generation of large amount of unstructured social media data. Thus, text data is being formed gradually in many areas like automated business, education, health care, show business. Opinion mining, the subfield of text mining, deals with mining of review text and classifying the opinions or the sentiments of that text as positive or negative. The work in this paper develops a framework for opinion mining. It includes a novel feature selection method called Most Persistent Feature Selection (MPFS). MPFS method uses information gain of the features in the review documents. The performance of the three different classifiers, namely Naïve Bayes, Maximum Entropy, and Support Vector Machine, with the proposed feature selection method is evaluated on movie reviews using the parameters accuracy, precision, recall, and F-score. The different classifier models generated show the acceptable performance in comparison with the other existing models.