Coordinated behavior detection within & across fully encrypted domains
Abstract:
Coordinated behavior and community detection play a critical role across a variety of disciplines, enabling the association of seemingly unrelated users with a single real-world group. The ability to link identities holds great promise for enhancing decision-making processes, user-oriented recommendation systems, entity resolution, the detection of frauds, money laundering, terrorist networks and state-actor influence campaigns. However, inconsistent data formats, variations in naming conventions, and malicious user behavior aimed at obscuring illicit activities hinder the accuracy of such endeavors. The complexity of coordinated activity detection is further amplified in fully-encrypted and anonymity preserving domains such as Blockchain and the dark web. Traditional techniques, which heavily rely on content analysis, face inherent limitations in these contexts. Other more sophisticated methods rely on domain-specific characteristics, or on interaction patterns controllable by the users themselves. As such, the active efforts of individuals to conceal their actions hinder accurate predictions within these domains. To address such challenges, we employ a network oriented approach, aiming to enhance community and coordinated activity detection techniques in the context of fully-encrypted domains. By shifting the focus to the underlying network structure and the temporal correlations between nodes’ activity, we exploit correlations in patterns which cannot be effectively controlled by the individuals within the system. We consider the Ethereum Blockchain as a designated environment for assessing our methodology and present encouraging results as to the ability of community and coordinated activity detection within and across fully encrypted domains.
Bio:
Dr. Shahar Somin, is currently a postdoctoral associate at Tel-Aviv University, under the joint guidance of Prof. Alex ‘Sandy’ Pentland (MIT) and Prof. Erez Shmueli. Shahar has held a research affiliate position at the MIT Media Lab since 2018 and is scheduled to officially commence a postdoctoral position at MIT in spring 2024. Currently, her research focuses on user identity and dynamics analytics within and across fully encrypted networks, with the primary objective of detecting illicit activities such as money laundering, terrorist networks, and state actor influence programs. Prior to her Ph.D. Shahar served as the head of Machine Learning research at the MIT-spinoff startup “Endor”. Shahar earned her Ph.D. from the Industrial Engineering department, in Tel-Aviv University, advised by Prof. Erez Shmueli. She also holds an M.Sc. degree in Computer Science and a B.Sc. degree in Mathematics and Computer Science, both obtained from the Hebrew University of Jerusalem.
