From Evidence to Trajectory: Abductive Reasoning Path Synthesis for Training Retrieval-Augmented Generation Agents

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Li, Muzhi, Qi, Jinhu, Wu, Yihong, Zhao, Minghao, Ma, Liheng, Li, Yifan, Wang, Xinyu, Zhang, Yingxue, Leung, Ho-fung, King, Irwin
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915519025643520
author Li, Muzhi
Qi, Jinhu
Wu, Yihong
Zhao, Minghao
Ma, Liheng
Li, Yifan
Wang, Xinyu
Zhang, Yingxue
Leung, Ho-fung
King, Irwin
author_facet Li, Muzhi
Qi, Jinhu
Wu, Yihong
Zhao, Minghao
Ma, Liheng
Li, Yifan
Wang, Xinyu
Zhang, Yingxue
Leung, Ho-fung
King, Irwin
contents Retrieval-augmented generation agents development is hindered by the lack of process-level supervision to effectively guide agentic capabilities like task decomposition, retriever invocation, and stepwise decision-making. While reinforcement learning offers a potential solution, it suffers from sparse rewards and the limited reasoning capabilities of large language models (LLMs). Meanwhile, existing data synthesis methods only produce chain-of-thought rationales and fail to model environmental interactions. In this paper, we propose EviPath, an evidence-anchored reasoning path synthesis paradigm for RAG agent development. EviPath comprises: (i) Abductive Subtask Planning, which decomposes the problem into sub-questions and iteratively plans an optimal solution path based on the dependencies between them; (ii) Faithful Sub-question Answering, which uses supporting evidence to construct a proxy environment to generate reasoning thoughts and answers for each sub-question; and (iii) Conversational Fine-Tuning, which formats the complete agent-environment interaction trajectory into a dialogue format suitable for Supervised Fine-Tuning. EviPath allows LLMs to learn complex reasoning and tool-use capabilities directly from synthesized data. Extensive experiments on widely-used question-answering benchmarks show that an 8B parameter model trained with EviPath-synthesized data significantly and consistently outperforms state-of-the-art baselines with a double-digit absolute EM gain of 14.7% in open-domain question answering.
format Preprint
id arxiv_https___arxiv_org_abs_2509_23071
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle From Evidence to Trajectory: Abductive Reasoning Path Synthesis for Training Retrieval-Augmented Generation Agents
Li, Muzhi
Qi, Jinhu
Wu, Yihong
Zhao, Minghao
Ma, Liheng
Li, Yifan
Wang, Xinyu
Zhang, Yingxue
Leung, Ho-fung
King, Irwin
Computation and Language
Artificial Intelligence
Retrieval-augmented generation agents development is hindered by the lack of process-level supervision to effectively guide agentic capabilities like task decomposition, retriever invocation, and stepwise decision-making. While reinforcement learning offers a potential solution, it suffers from sparse rewards and the limited reasoning capabilities of large language models (LLMs). Meanwhile, existing data synthesis methods only produce chain-of-thought rationales and fail to model environmental interactions. In this paper, we propose EviPath, an evidence-anchored reasoning path synthesis paradigm for RAG agent development. EviPath comprises: (i) Abductive Subtask Planning, which decomposes the problem into sub-questions and iteratively plans an optimal solution path based on the dependencies between them; (ii) Faithful Sub-question Answering, which uses supporting evidence to construct a proxy environment to generate reasoning thoughts and answers for each sub-question; and (iii) Conversational Fine-Tuning, which formats the complete agent-environment interaction trajectory into a dialogue format suitable for Supervised Fine-Tuning. EviPath allows LLMs to learn complex reasoning and tool-use capabilities directly from synthesized data. Extensive experiments on widely-used question-answering benchmarks show that an 8B parameter model trained with EviPath-synthesized data significantly and consistently outperforms state-of-the-art baselines with a double-digit absolute EM gain of 14.7% in open-domain question answering.
title From Evidence to Trajectory: Abductive Reasoning Path Synthesis for Training Retrieval-Augmented Generation Agents
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2509.23071