Abstract
Systems and methods maintain a per-user trust score for a recommendation system and adapt ranking behavior based on the trust score. The trust score is updated after recommendation outcomes using asymmetric dynamics in which positive outcomes increase trust with diminishing returns and negative outcomes decrease trust more strongly, optionally using severity weights for different negative signals and a small decay for neutral outcomes. The trust score is classified into discrete trust states that simultaneously modulate multiple ranking parameters, including a candidate quality threshold, an exploration rate, prediction-confidence weighting applied to ranking scores, and sensitivity to negative-signal predictions. When trust is low, a recovery mode may be entered that accelerates trust rebuilding based on positive streaks, biases serving toward a safe set, suppresses and then gradually reintroduces novelty, and exits based on trust and multi-session criteria.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Anonymous, "Systems and Methods for Trust-Adaptive Ranking with Asymmetric Recovery in Content Recommendation", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10755