Breaking the Filter Bubble: A Semantic Pareto-DQN Framework for Multi-Objective Recommendation

Short summary

Researchers introduce Semantic Pareto-DQN, a multi-objective reinforcement learning framework that balances user engagement, information diversity, and fairness in recommender systems. By treating these as distinct reward signals instead of aggregated metrics, the approach disrupts filter bubbles and semantic collapse. Empirical evaluation on MovieLens data demonstrates significant diversity gains with minimal trade-offs to user engagement.

• •Multi-objective RL framework addresses filter bubbles by balancing engagement, diversity, and fairness as distinct objectives
• •Pareto-DQN avoids semantic homogenization by maintaining high state-trajectory variance across reward signals
• •MovieLens experiments show diversity gains with negligible engagement loss

Generated with AI, which can make mistakes.