Terrorist Organization Identification Using Link Prediction over Heterogeneous GNN

Terrorist attacks are threats to undermine state security and citizens’ confidence. In the last year, the Member States of European Union reported a total of 57 completed, failed, and foiled terrorist attacks, with 21 people killed. Counter-terrorism activities, through intelligence analysis experts, attempts to face, tackle, and prevent new attacks. In this sense, Artificial Intelligence demonstrates promising support for data analysis and patterns identification in security concerns. This work treats the application of a deep learning approach for the association between attacks and perpetrator groups, which is often unknown. Identifying the most involved actors helps extract inherited features and better study attacks to implement suitable countermeasures and predict feature events. Starting from the well-known resource related to anti-terrorism operations, Global Terrorism Database (GTD), we build a knowledge graph (KG) representing entities and relationships involved in terrorist attacks. Subsequently, we adopted the KG to train a graph neural network (GNN) to identify terrorist organizations from events using the inductive link prediction technique. The experimentation, conducted by adopting the HinSAGE framework, demonstrates promising performance in terms of accuracy with a discrete improvement to state-of-the-art