filter-bubble

Proposes a multi-objective reinforcement learning framework combining semantic embeddings with Pareto-DQN to balance engagement, diversity, and fairness in recommendations, mitigating filter bubbles.