The field of recommender systems is moving towards addressing the issues of filter bubbles, echo chambers, and homogenization traps. Researchers are exploring innovative approaches to mitigate these biases, including the use of community detection algorithms, psychological mechanisms, and diversity-driven techniques. A key direction is the development of frameworks that can integrate multiple perspectives and promote normative diversity in news recommendations. Another important area of research is the improvement of negative sampling strategies to enhance the expressiveness of recommenders. Noteworthy papers in this area include: When Algorithms Mirror Minds, which proposes a Confirmation-Aware Social Dynamic Model to simulate the interaction between users and recommenders. D-RDW, which introduces a lightweight algorithm for generating diverse news recommendations. Democratizing News Recommenders, which proposes a framework for modeling multiple perspectives in news candidate generation. Diverse Negative Sampling, which explicitly accounts for diversity in negative training data during the negative sampling process.
Mitigating Biases in Recommender Systems
Sources
Mitigating Filter Bubble from the Perspective of Community Detection: A Universal Framework
When Algorithms Mirror Minds: A Confirmation-Aware Social Dynamic Model of Echo Chamber and Homogenization Traps
RRRA: Resampling and Reranking through a Retriever Adapter
Informfully Recommenders -- Reproducibility Framework for Diversity-aware Intra-session Recommendations
D-RDW: Diversity-Driven Random Walks for News Recommender Systems
Reactive Users vs. Recommendation Systems: An Adaptive Policy to Manage Opinion Drifts
Democratizing News Recommenders: Modeling Multiple Perspectives for News Candidate Generation with VQ-VAE
Diverse Negative Sampling for Implicit Collaborative Filtering
Benefiting from Negative yet Informative Feedback by Contrasting Opposing Sequential Patterns
Privacy Preserving Inference of Personalized Content for Out of Matrix Users