Enhancing the Detection of Social bots on Twitter using Ensemble machine Learning Technique

Abstract

Today our world experiences a large number of active social media users daily, twitter being one the most used platform for discussion on various topics like politics, sports, entertainment etc. It highly influences people's lives and therefore it is required to maintain a healthy environment in such places. Thus these places eventually become the epicenter of malicious activities, wherein someone tries to share hate or manipulate information as per their own interest. The common mass which comprises the most part of the user base having limited knowledge of such things, fall prey to these activities. At present millions of such automated accounts exist, also known as bots which are involved in malicious activities like spreading misinformation and manipulating public opinion. The work presented here is aimed at developing a framework by implementing ensemble machine learning approaches like Adaptive boosting, Gradient boost (GB) and Extreme Gradient boost (XGB) to detect these twitter bots. We have used a dataset that is publicly available from database community and evaluate our proposed approach to predict whether the user account is a bot or non-bot. Our experiment demonstrates that the estimator GB achieves highest accuracy in detecting the social bots.