PathRouter: Aligning Rewards with Retrieval Quality in Agentic Graph Retrieval-Augmented Generation
For researchers and practitioners using agentic retrieval-augmented generation with graph-structured data, PathRouter improves both answer accuracy and evidence quality, though the gains are incremental over strong baselines.
PathRouter addresses answer-path reward aliasing and search-update ambiguity in agentic GraphRAG by jointly evaluating trajectories on answer correctness and evidence-path overlap, using differentiated GRPO advantage scaling and token-level KL guidance. It achieves average F1 gains of 3.1 on 3B and 4.9 on 7B models across six QA benchmarks.
Agentic GraphRAG trains language-model agents to iteratively retrieve and reason over graph-structured evidence, enabling more accurate and context-aware decision-making by efficiently navigating complex information networks. However, outcome-only reinforcement learning suffers from \textit{\textbf{answer-path reward aliasing}}, where correct answers may come from shortcuts rather than useful evidence paths. It also exhibits \textit{\textbf{search-update ambiguity}}, as scalar trajectory-level feedback does not indicate which retrieval actions to adjust. To mitigate these shortcomings, we present PathRouter, a path-aware training framework for agentic GraphRAG. PathRouter jointly evaluates each trajectory along answer correctness and evidence-path overlap, yielding four trajectory categories with differentiated GRPO advantage scaling that suppresses shortcut reinforcement while preserving evidence-seeking behavior. For evidence-poor trajectories, a frozen gold-evidence teacher provides token-level KL guidance on reasoning and search-query tokens, excluding answer tokens to avoid direct response imitation. Experiments on six QA benchmarks across three model sizes show that PathRouter consistently improves answer F1 and evidence-path overlap, achieving average F1 gains of 3.1 on 3B and 4.9 on 7B models compared to a strong baseline.