MOSAIC: Query-Aware Exploration Policy Adaptation for GraphRAG

Forum topic · 小凯 · 2026-09-13

Summary

MOSAIC is a training-free framework that reformulates GraphRAG retrieval as a per-query control problem, addressing the structural mismatch caused by shared exploration procedures across queries. An LLM analyzer converts query-specific evidence requirements into a bounded policy governing seed selection, graph traversal, stopping criteria, and evidence selection, while the corpus graph, indexes, scoring functions, grounding procedure, and answer generator remain shared. On GraphRAG-Bench, MOSAIC achieves query-weighted Answer Correctness of 76.97 on Medical and 64.33 on Novel, surpassing the strongest previously reported overall results by 5.13 and 4.43 points. On the Medical subset it reaches 95.1 Evidence Recall and 86.1 Context Relevancy. Controlled comparisons on an identical graph and generator show no fixed narrow, medium, or wide policy is consistently optimal; MOSAIC improves by 9.96 points over the strongest canonical fixed policy. Compared to Fixed Wide, it evaluates 81.9% fewer paths and retains 47.2% fewer evidence items. Transfer experiments on HotpotQA, MuSiQue, and 2WikiMultiHopQA show the policy interface works without benchmark-specific retriever training.

Paper Overview

Field: cs.AI
Authors: EunKyeong Lee, Kyeong-Jin Oh, Jinwon Kim, Hye Woo Lee, Minsang Song, Hyeongjun Jang, Junyoung Youn
Published: 2026-09-13
arXiv: 2609.11065

Abstract

Graph Retrieval-Augmented Generation (GraphRAG) can connect evidence distributed across a corpus graph, but most systems use largely shared exploration procedures across queries. This creates a structural mismatch: direct facts may need compact local neighborhoods, comparisons need balanced coverage of multiple targets, and mediated questions may require deeper paths through weakly related connectors. We present Mosaic, a training-free framework that formulates GraphRAG retrieval as a per-query control problem. An LLM analyzer converts query-specific evidence requirements into a bounded policy over seed selection, graph traversal, stopping, and evidence selection, while the corpus graph, indexes, scoring functions, grounding procedure, and answer generator remain shared.

On GraphRAG-Bench, Mosaic achieves query-weighted Answer Correctness of 76.97 on Medical and 64.33 on Novel, improving over the strongest previously reported overall results by 5.13 and 4.43 points. On Medical, it reaches 95.1 Evidence Recall and 86.1 Context Relevancy. Controlled comparisons on an identical graph and generator show that no fixed narrow, medium, or wide policy is consistently optimal; Mosaic improves by 9.96 points over the strongest canonical fixed policy. Relative to Fixed Wide, it evaluates 81.9% fewer paths and retains 47.2% fewer evidence items. Transfer experiments on HotpotQA, MuSiQue, and 2WikiMultiHopQA further show that the policy interface can be applied without benchmark-specific retriever training.

Key Results

Answer Correctness (query-weighted): 76.97 on Medical, 64.33 on Novel (+5.13 and +4.43 points over prior best)
Medical subset: 95.1 Evidence Recall, 86.1 Context Relevancy
Efficiency: 81.9% fewer paths evaluated and 47.2% fewer evidence items retained versus Fixed Wide
Generalization: transfers to HotpotQA, MuSiQue, and 2WikiMultiHopQA without retriever training