Graph Neural Networks Based Framework to Analyze Social Media Platforms for Malicious User Detection

Abstract

Online Social Media (OSM) is 24/7 available to all user’s around the globe. User’s are seamlessly connected in an unstructured network leading to the seamless flow of information from east to west in fractions of seconds. The user’s on the social media are active with different intentions that could be malign intents or true later and spirit. Based on the user’s intentions, the contents being shared by end user’s can be defamed to spread misinformation, disinformation, propaganda and rumor. The ingress of such propagandistic contents in society can result in financial damage, panic, uncertainty and demoralizing the mob aimed to achieve any political and military objectives. In present era of internet and globalization, propaganda warfare is an integral part of 5th generation and hybrid warfare. Therefore, detection of user’s with malicious intentions is need of the hour aimed to stop spread of malicious contents into society. This paper proposed a deep learning based framework that exploits the social media in three different domains i.e, user’s profile, contents being shared and analysis of the user’s unstructured ego-network. The framework is established on inductive learning based neural network for 3D analysis of social media platform. The proposed model is kind of a benchmark in itself that can provide a base-line for the researchers. The performance of the proposed model is compared with already available  approaches i.e SVM and LSTM. Series of experiments renders the out performance of the proposed framework on real-world PHEME dataset. The proposed framework may also be used as an OSINT tool subject to availability of customized data.

Keywords: Social media unstructured data analytic, user's profile analysis, shared contents profiling, BERT, user-centered ego-network, malicious contents.