Towards a Comprehensive Approach for Socialbot Detection on Twitter: Integrating Multiple Features

Abstract

The increasing use of social media platforms has brought about various opportunities for people to communicate, share opinions, and express themselves online. However, this has also led to the emergence of socialbots, programmed accounts that mimic human behavior and have the potential to spread fake news, manipulate the stock market, promote terrorism, and interfere in democratic processes. In order to address this problem, various techniques have been employed for socialbots detection, including profile-based, temporal-based, content-based, behavioral-based, and network-based approaches. However, none of these methods have utilized a hybrid of all these features. In this paper, we propose a hybrid approach that integrates all these features to train a model for socialbots detection on Twitter. we use the Twibot-22 dataset for our experiments and evaluate the performance of our proposed approach against benchmark models. With an accuracy of 0.898, the XGBoost model surpassed the benchmark models. This study contributes to the ongoing efforts to maintain the integrity of tweet contents and address the potential harms caused by socialbots on social networks