Personal-Agent Mediated Recommendation with Cross-Platform User History
PAMO trains personal LLM agents to re-rank platform recommendations using cross-platform history with fewer harmful overrides.
The paper formalizes Personal-Agent Mediated Recommendation, where a platform ranks candidates from local data and a personal LLM agent uses authorized cross-platform history to produce the final top-K slate. MediateRec provides scalable proxy environments and a real cross-platform test under a controlled information boundary. Personal Attribution Mediation Optimization counterfactually masks user history to estimate mediation support and reallocates rank-aware advantage under a platform-relative value floor. Experiments show useful corrections, but even strong proprietary LLMs cause harmful overrides; PAMO improves over outcome-only RL on seen, unseen, and real tests.
- A personal agent re-ranks a platform slate using authorized cross-platform history.
- MediateRec tests proxy environments and a real cross-platform information boundary.
- PAMO estimates personal mediation support by counterfactually masking history.
- Strong proprietary LLMs still make non-negligible harmful overrides; PAMO improves the rescue-harm balance.
- Products
- MediateRecPAMO
From the original article
1 min read at huggingface.co
Modern recommendation is shifting from platform-centric personalization toward user-governed personalization, where a personal LLM agent can act on the user's behalf across services.
By Yu Xia, Jiangfan Zhang, Jun Xiao · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.