Analysis and detection of deviant and malicious behaviors in social media and beyond
Ανάλυση και εντοπισμός παραβατικών και κακόβουλων συμπεριφορών στα κοινωνικά δίκτυα και πέραν αυτών
View/
Keywords
Deviant behaviors ; Social media ; Adult content ; CybercrimeAbstract
Deviant and malicious behaviors are inevitably interwoven into the very fabric of the Internet, due to its ubiquitous nature and its inherent imperviousness to authoritative regulation and control. Such behaviors can potentially have a devastating impact on today’s hyper-connected world. The purpose of this thesis is to study several relatively unexplored facets of deviant and malicious online behaviors in the areas of social media and cybercrime. Concretely, this work is organized in two parts:
The first one focuses on the adult-content related deviant behaviors flourishing in the domain of Social Live Streaming Services (SLSS). This kind of social platforms allows a new level of social interaction, by enabling their users to share their daily lives through the cameras of their mobile devices. However, as they lack the mechanisms to effectively enforce their community guidelines, they are rife with adult content. This work examines in depth the mechanics of the adult production and consumption phenomenon in two large SLSS platforms in terms of interactions between their users and characterizing attributes of their behavior. Additionally, the largest-to-date dataset of chats and user interactions in the context of adult content live streams is constructed and analyzed to unveil evidence of sexual exploitation and grooming targeting underage users, as well as to disentangle the strategies adopted by malicious users to evade the moderation mechanisms of such platforms. Furthermore, this thesis sheds light on the semi-illicit adult content market layered on the top of popular social media platforms, its offerings, and the demographics of adult content producers.
The second part concentrates on the world of cybercrime. Specifically, this dissertation studies the modus operandi of cybercrime vendors who use anonymous marketplace platforms on the Surface Web to sell illicit products and services such as leaked credentials, breached accounts and malware, while hiding in plain sight. Particularly for the case of malware, this work delves into the problem of detecting algorithmically generated domains, an approach employed by modern malware and botnets to enhance and scale their persistence and orchestration capabilities over millions of infected devices. To this end, this work presents the largest-to-date dataset of such domains, and proposes a novel set of features useful for the resilient and robust detection of a wide set of domain generation algorithms through machine learning approaches. Finally, this thesis explores the novel threats of the emerging field of blockchain-based DNS alternatives, focusing on the Namecoin and Emercoin ecosystems which are found to be abused for malicious purposes.