Breaking Anonymity of Social Media by Profiling from Multimodal Information

Abstract

Information on social media is important for forensics in the real and cyber worlds, because social media platforms are used for various illicit purposes. However, the anonymity of social media accounts hinders their forensic use. Conventional methods of deanonymizing social media accounts are not sufficiently precise, because they only use information from words in social media posts. Another drawback is that those methods require not only information from the anonymous accounts that are the target of deanonymization, but also information from real-name accounts belonging to the same people who own the anonymous accounts. This paper proposes a new deanonymization method that profiles anonymous accounts to infer their users' attributes and matches the inferred attributes with the known attributes of candidates for deanonymization. In this method, the profiling of anonymous accounts uses not only word information but also information from posted sentences and images. Evaluation with data from 78 volunteers demonstrates the proposed method's viability.