Features Identification for Filtering Credible Content on Twitter Using Machine Learning Techniques

Abstract

In the present era of Internet, Twitter is one of the pivotal platforms for sharing the views and opinions of an individual related to any topic by means of tweets. However, the credibility of such tweets is unidentified. Demonetization is one of the events in India when lots of chaos happened among public, and people posted tweets whose authenticity was trailed by the question mark. In this study, we have discovered the credibility of user content on Twitter network with the help of 26 different features. The experiments have been carried out on more than 1 k user tweets related to demonization. For classifying the tweets into four different credibility classes (acceptable, somewhat acceptable, somewhat unacceptable, and unacceptable) several machine learning techniques such as random forest, Naive Bayes, and support vector machine has been utilized. Out of all the selected classifiers, random forest has been observed as the best classifier with accuracy and f1 score as 0.977 and 0.9911, respectively. Furthermore, out of 26 identified features, we have recognized 10 most distinctive features to efficiently distinguish the user tweets in different credibility classes.