Beyond Passive Preference Inference: User-Governed Preference Memory and Calibrated Initiative in Recommender Systems

Abstract

Recommender systems remain organized around passive preference inference: estimate latent interest from behavioral traces, then rank. Past: implicit feedback, matrix factorization, learning-to-rank, and large-scale offline evaluation made this abstraction highly productive. Present: four mismatches expose what prediction alone cannot settle—under-specified goals, evidence shifting from sparse clicks to natural-language preference statements, a wider action space than ranking, and evaluation that lags proactive behavior. Future: the next decade should center two research objects, user-governed preference memory and calibrated initiative, composed inside a user-side or hybrid architecture and audited by agency-centered evaluation that adds timing quality, memory correctness, and revocability to accuracy. Within two RecSys cycles, we ask the community to develop a shared reference schema, a mixed-initiative benchmark suite re-annotated from conversational-recommendation corpora, and a PPF-track reporting convention. The supplement contributes the machine-readable PreferenceMemoryRecord v0 JSON Schema. This agenda complements ranking-centric work with a common design surface.