METRO: Towards Strategy Induction from Expert Dialogue Transcripts for Non-collaborative Dialogues

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Yang, Haofu, Liu, Jiaji, Huang, Chen, Wu, Faguo, Lei, Wenqiang, Ng, See-Kiong
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866915940168368128
author Yang, Haofu
Liu, Jiaji
Huang, Chen
Wu, Faguo
Lei, Wenqiang
Ng, See-Kiong
author_facet Yang, Haofu
Liu, Jiaji
Huang, Chen
Wu, Faguo
Lei, Wenqiang
Ng, See-Kiong
contents Developing non-collaborative dialogue agents traditionally requires the manual, unscalable codification of expert strategies. We propose \ours, a method that leverages large language models to autonomously induce both strategy actions and planning logic directly from raw transcripts. METRO formalizes expert knowledge into a Strategy Forest, a hierarchical structure that captures both short-term responses (nodes) and long-term strategic foresight (branches). Experimental results across two benchmarks show that METRO demonstrates promising performance, outperforming existing methods by an average of 9%-10%. Our further analysis not only reveals the success behind METRO (strategic behavioral diversity and foresight), but also demonstrates its robust cross-task transferability. This offers new insights into building non-collaborative agents in a cost-effective and scalable way. Our code is available at https://github.com/Humphrey-0125/METRO.
format Preprint
id arxiv_https___arxiv_org_abs_2604_11427
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle METRO: Towards Strategy Induction from Expert Dialogue Transcripts for Non-collaborative Dialogues
Yang, Haofu
Liu, Jiaji
Huang, Chen
Wu, Faguo
Lei, Wenqiang
Ng, See-Kiong
Computation and Language
Artificial Intelligence
Developing non-collaborative dialogue agents traditionally requires the manual, unscalable codification of expert strategies. We propose \ours, a method that leverages large language models to autonomously induce both strategy actions and planning logic directly from raw transcripts. METRO formalizes expert knowledge into a Strategy Forest, a hierarchical structure that captures both short-term responses (nodes) and long-term strategic foresight (branches). Experimental results across two benchmarks show that METRO demonstrates promising performance, outperforming existing methods by an average of 9%-10%. Our further analysis not only reveals the success behind METRO (strategic behavioral diversity and foresight), but also demonstrates its robust cross-task transferability. This offers new insights into building non-collaborative agents in a cost-effective and scalable way. Our code is available at https://github.com/Humphrey-0125/METRO.
title METRO: Towards Strategy Induction from Expert Dialogue Transcripts for Non-collaborative Dialogues
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2604.11427