Designing Technological Solutions for Trust and Safety in Online Social Platforms

The rapid diffusion of online social networks, messaging platforms, and social virtual reality (VR) environment has transformed global communication while raising pressing concerns about privacy, misinformation, and online safety. This thesis addresses these challenges by designing and evaluating technological solutions that enhance user protection, empower individuals, and promote inclusive digital environments. First, we investigate content moderation in online social networks, integrating large language models (LLMs) into moderation pipelines to overcome rigidity and opacity of existing systems. Experiments with selected LLMs on real-world datasets demonstrate that these models are capable of providing personalized moderation and natural-language explanations, although they are not yet suitable for use as standalone tools. Second, we propose a human-AI pipeline for content moderation in social VR, specifically aimed at addressing harassment through a combination of technical countermeasures and adaptive strategies designed to meet the unique challenges of immersive, three-dimensional environments. Third, we examine misinformation prevention, demonstrating that LLMs can detect misleading content and assist users in avoiding its accidental dissemination. We introduce user-centered design patterns, such as inline suggestions and real-time feedback, which not only help curb the unintentional spread of misinformation but also contribute to shifting the focus of moderation from platform-driven enforcement toward user empowerment. Fourth, we address the non-consensual sharing of intimate images in messaging applications. Using Telegram as a case study, we propose the development of a tool to map abusive networks through graph analysis and perceptual hashing. We also design SafeSext, a proof-of-concept messaging system that prevent unauthorized forwarding through the combined use of nudity detection and fingerprinting techniques, and propose decentralized extensions using blockchain and NFTs to strengthen protections across platforms. Finally, we develop Safe Digital Teens, an educational application that teaches adolescents safe online behaviors. A user study confirms its usability and highlight its potential for digital citizenship training. By combining AI-driven moderation, system design, and educational tools, this thesis contributes to safer, fairer, and more transparent digital platforms. Our findings advance multiple UN Sustainable Development Goals, underscoring the societal importance of trustworthy online spaces.