Can We Trust This User? Predicting Insider's Attitude via YouTube Usage Profiling

摘要

Addressing the insider threat is a major issue in cyber and corporate security in order to enhance trusted computing in critical infrastructures. In this paper we study the psychosocial perspective and the implications of insider threat prediction via social media, Open Source Intelligence and user generated content classification. Inductively, we propose a prediction method by evaluating the predisposition towards law enforcement and authorities, a personal psychosocial trait closely connected to the manifestation of malevolent insiders. We propose a methodology to detect users holding negative attitude towards authorities. For doing so, we facilitate a brief analysis of the medium (YouTube), machine learning techniques and a dictionary-based approach, in order to detect comments expressing negative attitude. Thus, we can draw conclusions over a user behavior and beliefs via the content the user generated within the limits a social medium. We also use an assumption free flat data representation technique in order to decide over the user's attitude and improve the scalability of our method. Furthermore, we compare the results of each method and highlight the common behavior and characteristics manifested by the users. As privacy violations may well-rise when using such methods, their use should be restricted only on exceptional cases, e.g. when appointing security officers or decision-making staff in critical infrastructures.