Behavior Based Machine Learning Approaches to Identify State-Sponsored Trolls on Twitter

Abstract

Social media is a powerful weapon that youth is much interested in due to the wider availability of internet in the recent times. Due to its wider availability, there is a need and necessity to keep track of the activity or interests that youth is much interested upon based on the individual activity to ensure that he is behaving right, that pays a new way of research in this field. For this case we have considered Twitter, social media site that a larger user base to conduct our study where the user data is retrieved and scrutinized based on the preferred language, tweets and preferences by other users and their social behavior etc. are considered for data analysis to understand various kinds of texts in different languages. The proposed work is further enhanced with the use of machine learning algorithms to determine the persons emotions and to eliminate the fake statements and estimate the kind of tweets whether considered as good or bad. Based on the provided training and test data using different algorithms based on machine learning such as SVM, Naïve Bayes and LDA algorithms the personality is analyzed firstly, later the use of Random Forest and extra gradient boosting techniques by connecting tweety through an API can determine the truthfulness of the statement provided by the individual achieving a greater accuracy of 90 and 99.61 percent accordingly for the estimation of performance analysis.