Advancing Social Network Analytics: Resilience and Security

In the digital age, Online Social Networks (OSNs) have emerged as epicenters of human interaction, facilitating the creation, sharing, and dissemination of information at an unprecedented scale. The vast reservoir of user-generated data within OSNs has become a valuable resource for researchers, analysts, and practitioners. Social Network Analytics (SNA) has arisen as a powerful tool to extract insights from these data, enabling a better understanding of social structures and dynamics. However, the ever-changing landscape of OSNs, marked by the emergence of new platforms and shifts in user behavior, necessitates constant adaptation of SNA methodologies and tools. This dissertation advances SNA in three dimensions: (i) explaining influence and engagement mechanisms in trending OSNs; (ii) developing resilient SNA tools designed to function effectively in adversarial environments, and (iii) exploring security and privacy concerns in modern social platforms. The first part of this thesis begins by examining how virtual influencers are transforming OSNs and influencer marketing. While major companies and brands increasingly embrace them, individuals remain divided between enthusiasm and apprehension. The thesis then unveils Instagram engagement mechanisms to optimize content creation and delves into TikTok’s unique influence dynamics, emphasizing how influencers can expand their reach. These studies demonstrate that influence and engagement patterns are strictly related to the tiers and categories of influencers, an aspect not considered in the existing literature. The part concludes with a case study exemplifying information manipulation on Twitter orchestrated by social bots. Motivated by the existence of such adversarial activities, the second part of the thesis focuses on developing resilient SNA tools tailored for adversarial contexts. The part begins by identifying Instagram crowdturfing, an emerging phenomenon wherein individuals generate fake engagement using their authentic profiles, behind a monetary reward. The analysis reveals that over 20% of influencers’ engagement is artificial. Then, the thesis delves into categorizing simple but powerful obfuscation techniques OSN users adopt to evade moderators, and proposes a detection mechanism based on supervised and unsupervised Deep Learning (DL) strategies. The part concludes by introducing the innovative concept of social honeypots for examining OSN communities and trends. These honeypots are fully automated Instagram pages, powered by generative AI, that attract users for subsequent analysis. Eventually, the notion of social networks has expanded beyond traditional OSNs to encompass contemporary digital landscapes like video games and the Metaverse. In fact, within these virtual worlds, people engage, communicate, and forge connections, giving rise to online communities and social interactions. However, the widespread use of these modern social platforms also results in the generation of massive amounts of (public) data, which can be exploited for malicious purposes. Unfortunately, numerous threats within this evolving landscape remain unknown, while techniques to identify malicious users can be the key to mitigating these risks. This thesis’s final part focuses on enhancing security and privacy in modern social platforms, such as video games and the Metaverse. First, it introduces PvP, a DL-based framework that can effectively identify gamers based on their play style. Then, the thesis assesses an attribute inference attack in Dota 2, with far-reaching privacy consequences for millions of gamers. In fact, it demonstrates that players’ private information, including their age, gender, or personality traits, can be inferred with up to 96% precision. The thesis concludes by presenting a comprehensive user profiling framework for augmented and virtual reality, addressing privacy and security challenges within the Metaverse’s enabling technologies.