A Methodology For Large-Scale Identification of Related Accounts in Underground Forums

Abstract

Underground forums allow users to interact with communities focused on illicit activities. They serve as an entry point for actors interested in deviant and criminal topics. Due to the pseudo-anonymity provided, they have become improvised marketplaces for trading illegal products and services, including those used to conduct cyberattacks. Thus, these forums are an important data source for threat intelligence analysts and law enforcement. The use of multiple accounts is forbidden in most forums since these are mostly used for malicious purposes. Still, this is a common practice. Being able to identify an actor or gang behind multiple accounts allows for proper attribution in online investigations, and also to design intervention mechanisms for illegal activities. Existing solutions for multi-account detection either require ground truth data to conduct supervised classification or use manual approaches. In this work, we propose a methodology for the large-scale identification of related accounts in underground forums. These accounts are similar according to the distinctive content posted, and thus are likely to belong to the same actor or group. The methodology applies to various domains and leverages distinctive artefacts and personal information left online by the users. We provide experimental results on a large dataset comprising more than 1.1M user accounts from 15 different forums. We show how this methodology, combined with existing approaches commonly used in social media forensics, can assist with and improve online investigations.