UP-NRPA: User Portrait based Nested Rollout Policy Adaptation for Planning with Large Language Models in Goal-oriented Dialogue Systems

In three linesUP-NRPA, a user portrait-based framework, dynamically adapts dialogue strategies with LLMs without offline reinforcement learning. On collaborative and non-collaborative benchmarks, the system achieves 100% success rate across multiple tasks and increases sale-to-list ratio by 56.41% in negotiation.

Summary generated by Claude — human-verified