Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GraphWalker: Agentic Knowledge Graph Question Answering via Synthetic Trajectory Curriculum

Shuwen Xu, Yao Xu, Jiaxiang Liu, Chenhao Yuan, Wenshuo Peng, Jun Zhao, Kang Liu

Abstract

Agentic knowledge graph question answering (KGQA) requires an agent to iteratively interact with knowledge graphs (KGs), posing challenges in both training data scarcity and reasoning generalization. Specifically, existing approaches often restrict agent exploration: prompting-based methods lack autonomous navigation training, while current training pipelines usually confine reasoning to predefined trajectories. To this end, this paper proposes \textit{GraphWalker}, a novel agentic KGQA framework that addresses these challenges through \textit{Automated Trajectory Synthesis} and \textit{Stage-wise Fine-tuning}. GraphWalker adopts a two-stage SFT training paradigm: First, the agent is trained on structurally diverse trajectories synthesized from constrained random-walk paths, establishing a broad exploration prior over the KG; Second, the agent is further fine-tuned on a small set of expert trajectories to develop reflection and error recovery capabilities. Extensive experiments demonstrate that our stage-wise SFT paradigm unlocks a higher performance ceiling for a lightweight reinforcement learning (RL) stage, enabling GraphWalker to achieve state-of-the-art performance on CWQ and WebQSP. Additional results on GrailQA and our constructed GraphWalkerBench confirm that GraphWalker enhances generalization to out-of-distribution reasoning paths. The code is publicly available at https://github.com/XuShuwenn/GraphWalker

GraphWalker: Agentic Knowledge Graph Question Answering via Synthetic Trajectory Curriculum

Abstract

Agentic knowledge graph question answering (KGQA) requires an agent to iteratively interact with knowledge graphs (KGs), posing challenges in both training data scarcity and reasoning generalization. Specifically, existing approaches often restrict agent exploration: prompting-based methods lack autonomous navigation training, while current training pipelines usually confine reasoning to predefined trajectories. To this end, this paper proposes \textit{GraphWalker}, a novel agentic KGQA framework that addresses these challenges through \textit{Automated Trajectory Synthesis} and \textit{Stage-wise Fine-tuning}. GraphWalker adopts a two-stage SFT training paradigm: First, the agent is trained on structurally diverse trajectories synthesized from constrained random-walk paths, establishing a broad exploration prior over the KG; Second, the agent is further fine-tuned on a small set of expert trajectories to develop reflection and error recovery capabilities. Extensive experiments demonstrate that our stage-wise SFT paradigm unlocks a higher performance ceiling for a lightweight reinforcement learning (RL) stage, enabling GraphWalker to achieve state-of-the-art performance on CWQ and WebQSP. Additional results on GrailQA and our constructed GraphWalkerBench confirm that GraphWalker enhances generalization to out-of-distribution reasoning paths. The code is publicly available at https://github.com/XuShuwenn/GraphWalker

Paper Structure

This paper contains 35 sections, 4 equations, 9 figures, 9 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. GraphWalker Framework
  3. Agentic Initialization
  4. Data Construction
  5. GraphSynth: Diverse Trajectory Synthesis
  6. GraphRoll: Expert Trajectory Collection
  7. GraphWalker Training
  8. Experiments
  9. Experimental Setup
  10. Main Results (RQ1)
  11. Comparison Analysis (RQ2)
  12. Analysis on Each Training Stage (RQ3)
  13. Ablation Study (RQ4)
  14. Related work
  15. Conclusion
  16. ...and 20 more sections

Figures (9)

  • Figure 1: Comparison of three KGQA paradigms.
  • Figure 2: Overview of our proposed GraphWalker.(a) Data Construction: a multi-phase pipeline for GraphSynth-15k and outcome-based rejection sampling for GraphRoll-6k. (b) Two-Stage SFT: Stage 1 establishes a broad exploration prior, while Stage 2 instills reflection and error recovery. (c) RL Optimization: GRPO with a simple exact-match reward unlocks a higher performance ceiling enabled by the two-stage SFT prior.
  • Figure 3: Representative interaction paths generated by constrained random walks.
  • Figure 4: Pass@$k$ comparison.
  • Figure 5: Ablation study on the quality filter in GraphSynth Phase 2.
  • ...and 4 more figures