A per-instance gated alignment between biased and unbiased ranking towers improved unbiased recommendation accuracy and, in live deployment at Kuaishou, slightly raised retention and interest diversity with negligible latency cost.
Feedback Loop and Bias Amplification in Recommender Systems
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
abstract
Recommendation algorithms are known to suffer from popularity bias; a few popular items are recommended frequently while the majority of other items are ignored. These recommendations are then consumed by the users, their reaction will be logged and added to the system: what is generally known as a feedback loop. In this paper, we propose a method for simulating the users interaction with the recommenders in an offline setting and study the impact of feedback loop on the popularity bias amplification of several recommendation algorithms. We then show how this bias amplification leads to several other problems such as declining the aggregate diversity, shifting the representation of users' taste over time and also homogenization of the users experience. In particular, we show that the impact of feedback loop is generally stronger for the users who belong to the minority group.
fields
cs.IR 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
ConAlign: Conditional Alignment Framework for Balancing Biased and Unbiased Recommendation
A per-instance gated alignment between biased and unbiased ranking towers improved unbiased recommendation accuracy and, in live deployment at Kuaishou, slightly raised retention and interest diversity with negligible latency cost.