Abstract
Systems and methods are disclosed for managing negative preference boundaries in content recommendation. A boundary record is identified from explicit signals (e.g., natural-language dislikes, hides) or implicit signals (e.g., repeated exposure with low engagement) and stored with a confidence score and timestamps. Boundary confidence decays over time using an exponential decay with a floor. When a boundary’s decayed confidence meets a testing criterion, the system generates semantically adjacent challenge preferences using an ontology, including intersection bridges with liked topics, softened variants, or meta-content framings, and selects bridges using a plausibility score. One or more probe items matching a selected challenge preference are injected into mid-slate positions and tagged for tracking. User responses are classified as positive, negative, or neutral, and boundary confidence and scheduling are updated accordingly, including exponential backoff after negative responses and weakening after positive responses. Natural-language profiles may be refined to qualify boundaries based on probe outcomes.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Anonymous, "Challenge Preference Generation and Boundary Testing for Filter Bubble Prevention in Content Recommendation Systems", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10723