• algorithmic ranking
• Artificial Intelligence Act (AI Act)
• automated decision-making
• deep-learning
• Digital Services Act (DSA)
• diversity
• electoral integrity
• engagement optimization
• Facebook
• General Data Protection Regulation (GDPR)
• Instagram
• Meta
• personalization loops
• profiling
• systemic-risk
• very large online platforms (VLOPs)

This dissertation shows how Meta’s deep-learning recommenders for Facebook and Instagram reorganize attention through engagement-optimized personalization loops and feedback dynamics, systematically privileging high-arousal content and opacity. It argues these architectures act as de facto gatekeepers with documented effects on freedom of expression and information pluralism, equality and non-discrimination, privacy and data protection, and democratic participation. Assessing the EU’s legal triad, the thesis finds the DSA, GDPR, and AI Act necessary but jointly insufficient. Transparency is often narrative, systemic-risk and audit duties lack shared metrics, the AI Act’s sectoral scoping misses platform-wide recommenders, and GDPR safeguards map imperfectly onto exposure-shaping profiling. It proposes operational fixes: evidence-bearing “Recommender System Cards,” standardized outcome metrics (e.g., exposure diversity, group-differential impacts), truly independent audits with log-level access, privacy-preserving researcher access under DSA Art. 40, stronger defaults in high-risk contexts (minors/elections), and coordinated DSA–GDPR–AI-Act guidance.