SimRec: Learning to Search, Ask, and Recommend from Simulated Interaction
arXiv:2611.15426v1 Announce Type: new Abstract: LLM-based recommendation agents can search catalogs, inspect products, ask users, and revise rejected recommendations. However, these behaviors are often governed by hand-designed workflows, while existing training methods optimize isolated components such as query rewriting or list ranking rather than the interaction policy as a whole. We present SimRec, a reinforcement learning framework that trains the full interaction policy of a tool-integrate
阅读原始来源
来源记录
项目分析
暂无五维分析，可先查看上方来源资料。