Abstract
Techniques are described for partitioning user profiles in agentic recommendation systems into a preference partition and a behavioral vulnerability partition with differentiated access control. A ranking agent retrieves only the preference partition to rank candidate content items, while a separate well-being monitoring system retrieves only the behavioral partition to compute a vulnerability-related measure and determine interventions. The well-being monitoring system transmits serving constraints to the ranking agent via a one-way constraint interface that excludes behavioral vulnerability indicators, such that delivery behavior may be adjusted (e.g., pacing, breaks, item limits, resource surfacing) without exposing behavioral vulnerability data to the ranking objective. Partitioning may be performed by classifying engagement-derived signals and decomposing ambiguous signals into preference and behavioral components routed to different partitions. Separate storage and retrieval APIs may enforce the access boundary.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Anonymous, "Partitioned User Profile Architecture with Differentiated Access Control for Recommendation Systems", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10751