Existence Conditions for Hidden Feedback Loops in Online Recommender Systems

We explore a hidden feedback loops effect in online recommender systems. Feedback result degradation of multi-armed bandit (MAB) recommendations to small subset and loss coverage novelty. study how uncertainty noise user interests influence the existence loops. First, we show that an unbiased additive random does not prevent loop. Second, demonstrate non-zero probability resetting is sufficient limit loop estimate size effect. Our experiments confirm theoretical findings simulated environment for four algorithms.