A2RAG: Adaptive Agentic Graph Retrieval for Cost-Aware and Reliable Reasoning

TL;DR

A2RAG introduces an adaptive, agentic framework for graph retrieval in multihop question answering, improving accuracy and efficiency by selectively refining evidence and maintaining robustness against extraction loss.

Contribution

It proposes a novel adaptive controller and agentic retriever that optimize retrieval efforts and ensure reliable reasoning in graph-based QA systems.

Findings

Achieves +9.9/+11.8 in Recall@2 on HotpotQA and 2WikiMultiHopQA.

Reduces token consumption and latency by about 50%.

Demonstrates improved cost-effectiveness and robustness in multihop QA.

Abstract

Graph Retrieval-Augmented Generation (Graph-RAG) enhances multihop question answering by organizing corpora into knowledge graphs and routing evidence through relational structure. However, practical deployments face two persistent bottlenecks: (i) mixed-difficulty workloads where one-size-fits-all retrieval either wastes cost on easy queries or fails on hard multihop cases, and (ii) extraction loss, where graph abstraction omits fine-grained qualifiers that remain only in source text. We present A2RAG, an adaptive-and-agentic GraphRAG framework for cost-aware and reliable reasoning. A2RAG couples an adaptive controller that verifies evidence sufficiency and triggers targeted refinement only when necessary, with an agentic retriever that progressively escalates retrieval effort and maps graph signals back to provenance text to remain robust under extraction loss and incomplete graphs. Experiments on HotpotQA and 2WikiMultiHopQA demonstrate that A2RAG achieves +9.9/+11.8 absolute gains in Recall@2, while cutting token consumption and end-to-end latency by about 50% relative to iterative multihop baselines.