Is a recommendation feed discovering your preferences or helping shape them?
A system can learn from what you click, while also deciding which options you get to see. That creates a feedback question: how much does the next round of data reflect existing preferences, and how much reflects earlier recommendations? A simulation study makes the loop explicit, without establishing how every commercial feed behaves.
Questions you could explore
- What data would help separate preference from exposure?
- Which choices never become visible in the click record?
Start reading
- arXiv / Chaney, Stewart, and EngelhardtHow Algorithmic Confounding in Recommendation Systems Increases Homogeneity and Decreases Utility
The abstract describes simulations in which recommendations affect the behavior later used as data. It cannot establish what happens on every real platform.
Ready to try your ideas aloud?
Your topic carries over. Record up to 2½ minutes, depending on your plan.