Social Media Tri-Domain Analysis for Detection of Potential/Likely Malicious Users

Online social media(OSM) is at the doorstep of every individual in this era of the global village. According to a recent study by the UN census bureau, 59% world’s population is actively using it, and 7 users are added to this figure every second. Users form an unstructured network on these platforms. Thus contents posted by any user would transverse in fractions of seconds. Contents available on social media platform effectively exploits the opinions of the readers. These contents can be defamed to spread misinformation, disinformation, and propaganda. The Spread of such propagandist content in society can adversely lead to fear, uncertainty, panic, or even financial loss in trading markets. Therefore, the detection of those users who spread such propagandistic content is the need of the hour. The users and their interactions with each other can be anticipated as similar to nodes and edges of the graph data structures. The edges are to be non-directed if interactions are two way like Facebook and directed if it is one way like Twitter. In this research, objective is to propose a graph convolution neural (GCN) network-based framework that captures the insights of the online social media unstructured patterns. The user-user request/ response graph is learned by the proposed framework. We suggest using the historic features of users to formulate the user profile. The performance of the proposed model is compared with SVM and LSTM. A series of experiments render the out-performance of the proposed framework on a real-world PHEME dataset. The proposed framework may also be used as an OSINT tool if customized data is available.