Hybrid Method of Information Gain and Particle Swarm Optimization for Selection of Features of SVM-Based Sentiment Analysis

Abstract

Sentiment analysis or opinion mining is a process of classifying opinions, usually from a text, toward a particular issue, to be positive, negative, or neutral. Nowadays, due to high number of social media users such as twitter, the opinion of social media users are often used to determine the public opinion. This can be used to find out the response of users to a particular candidate in an election or even to predict the election results. One of the challenges of using a document from social media, such as tweets, is the high number of attributes used in comparison to the length of documents, which is usually very short. In addition, the users tend to use informal languages in their tweets. In this paper, we propose to use particle swarm optimization (PSO) and Information Gain to select most appropriate attributes from documents and use support vector machine (SVM) as the classifier. We develop a sentiment analysis system for the election of West Java Governor. The experimental results show that our proposed system achieve the accuracy of 94.80% and area-under-curve (AUC) value of 0.98. Our results also show large improvements are achieved as the results of using PSO dan Information gain compared to without using them.