BriefGPT.xyz
May, 2024
通过LLM支持的用户代理模拟增强对话状态跟踪模型
Enhancing Dialogue State Tracking Models through LLM-backed User-Agents Simulation
HTML
PDF
Cheng Niu, Xingguang Wang, Xuxin Cheng, Juntong Song, Tong Zhang
TL;DR
利用GPT-4生成对话数据,通过在LLaMA 2上进行两阶段的微调,减少对话收集和注释成本,并表现出比仅使用真实数据训练的基准模型更好的性能,同时适应实际场景中的动态需求。
Abstract
dialogue state tracking
(DST) is designed to monitor the evolving dialogue state in the conversations and plays a pivotal role in developing
task-oriented dialogue systems
. However, obtaining the annotated data f
→