Abstract
In media recommendation architectures, engagement metrics such as comment participation probabilities are utilized to rank content. A limitation exists where media items with intentionally disabled comments are unfairly demoted. In these instances, a lack of engagement is processed as a lack of user interest. To address this deficiency, a boolean feature indicating a disabled comment status is generated. This boolean feature is subsequently crossed with a comment participation probability feature. Through this mathematical cross operation, a separate weight is established for scenarios where comments are disabled. Zero-engagement values are thereby processed differently when comment functionalities are unavailable. Consequently, engagement bias is mitigated, ensuring that media items are evaluated accurately without being structurally penalized for the utilization of platform safety mechanisms.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Ary, Acar; Sun, Jason; and Patwari, Ayush, "Mitigating Engagement Bias in Recommendation Architectures for Media Platforms", Technical Disclosure Commons, (June 30, 2026)
https://www.tdcommons.org/dpubs_series/10665