Advancing the Trustworthiness and Effectiveness of Recommender Systems

Every day, we rely on recommenders for movies, news, and shopping, but these systems often prioritize clicks over our safety and true interests. This thesis builds a blueprint for trustworthy AI by tackling four main challenges. First, it improves performance by filtering out noise to discover your genuine preferences. Second, it enhances adaptability, helping recommenders quickly understand new users with minimal data. Third, it exposes how hackers steal personal data from AI outputs, helping to build stronger privacy locks. Finally, it trains algorithms to detect and block evolving online scams, making everyday recommenders smarter and safer.