Influence factor based opinion mining of Twitter data using supervised learning

Social Networking portals have been widely used for expressing opinions in the public domain through internet based text messages and images. Among popular social networking portals, Twitter has been the point of attraction to several researchers in important areas like prediction of democratic electoral events, consumer brands, movie box office, stock market, popularity of celebrities etc. Sentiment analysis over Twitter offers a fast and efficient way of monitoring the public sentiment. In this paper, we introduce the novel approach of exploiting the user influence factor in order to predict the outcome of an election result. We also propose a hybrid approach of extracting opinion using direct and indirect features of Twitter data based on Support Vector Machines (SVM), Naive Bayes, Maximum Entropy and Artificial Neural Networks based supervised classifiers. We combined Principal Component Analysis (PCA) with SVM in an attempt to perform dimensionality reduction. This paper shows two different case studies of entirely different social scenarios, US Presidential Elections 2012 and Karnataka Assembly Elections 2013. We conclude the conditions under which Twitter may fail or succeed in predicting the outcome of elections. Experimental results demonstrate that Support Vector Machines outperform all other classifiers with maximum successful prediction accuracy of 88% in case of US Presidential Elections held in November 2012 and maximum prediction accuracy of 58% in case of Karnataka State Assembly Elections held in May 2013.