RouteRAG: Efficient Retrieval-Augmented Generation from Text and Graph via Reinforcement Learning

TL;DR

RouteRAG introduces an RL-based framework for adaptive, multi-turn hybrid retrieval-augmented generation from text and graphs, improving reasoning efficiency and accuracy in complex question answering tasks.

Contribution

The paper presents RouteRAG, a novel RL-based approach that enables end-to-end adaptive retrieval and reasoning from both text and graph data sources in RAG systems.

Findings

Significantly outperforms existing RAG baselines on five QA benchmarks.

Effectively balances retrieval efficiency and reasoning accuracy.

Supports multi-turn, adaptive reasoning with hybrid evidence.

Abstract

Retrieval-Augmented Generation (RAG) integrates non-parametric knowledge into Large Language Models (LLMs), typically from unstructured texts and structured graphs. While recent progress has advanced text-based RAG to multi-turn reasoning through Reinforcement Learning (RL), extending these advances to hybrid retrieval introduces additional challenges. Existing graph-based or hybrid systems typically depend on fixed or handcrafted retrieval pipelines, lacking the ability to integrate supplementary evidence as reasoning unfolds. Besides, while graph evidence provides relational structures crucial for multi-hop reasoning, it is substantially more expensive to retrieve. To address these limitations, we introduce \model{}, an RL-based framework that enables LLMs to perform multi-turn and adaptive graph-text hybrid RAG. \model{} jointly optimizes the entire generation process via RL, allowing the model to learn when to reason, what to retrieve from either texts or graphs, and when to produce final answers, all within a unified generation policy. To guide this learning process, we design a two-stage training framework that accounts for both task outcome and retrieval efficiency, enabling the model to exploit hybrid evidence while avoiding unnecessary retrieval overhead. Experimental results across five question answering benchmarks demonstrate that \model{} significantly outperforms existing RAG baselines, highlighting the benefits of end-to-end RL in supporting adaptive and efficient retrieval for complex reasoning.