Disentanglement and Interpretability in Recommender Systems
From Data Skeptic by Kyle Polich
March 10, 2026 · 31 min
About this episode
Ervin Dervishaj discusses his research on disentangled representation learning in recommender systems and its implications for interpretability and user trust.
Ervin Dervishaj, a PhD student at the University of Copenhagen, discusses his research on disentangled representation learning in recommender systems, finding that while disentanglement strongly correlates with interpretability, it doesn't consistently improve recommendation performance. The conversation explores how disentanglement acts as a regularizer that can enhance user trust and interpretability at the potential cost of some accuracy, and touches on the future of large language models in denoising user interaction data.
People in this episode
Host: Kyle Polich
Guest: Ervin Dervishaj
Topics covered
- disentangled representation learning
- recommender systems
- interpretability
- user trust
- large language models
Keywords
- disentanglement
- interpretability
- recommender systems
- user trust
- large language models
- accuracy
- regularizer
Mentioned in this episode
Organizations: University of Copenhagen
More episodes of Data Skeptic
- Student Spotlight: Aaron Payne, Data Analyst · May 1, 2026 · 26 min
- The Future is Agentic in Recommender Systems · April 25, 2026 · 49 min
- Book Ratings and Recommendations · March 27, 2026 · 39 min
- Collective Altruism in Recommender Systems · February 27, 2026 · 55 min
- Niche vs Mainstream · February 18, 2026 · 34 min
- Healthy Friction in Job Recommender Systems · February 2, 2026 · 27 min
Explore listener stats, chart rankings, contacts and more on the Data Skeptic podcast page.