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. Instead of applying one shared exploration procedure to all queries, an LLM analyzer converts query-specific evidence requirements into a bounded policy covering seed selection, graph traversal, stopping, and evidence selection, while the corpus graph, indexes, scoring functions, grounding, and answer generator remain shared. This addresses the structural mismatch between query types: direct facts need compact local neighborhoods, comparisons require balanced multi-target coverage, and mediated questions demand deeper paths through weakly related connectors. 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 show no fixed narrow, medium, or wide policy is consistently optimal; MOSAIC improves 9.96 points over the strongest canonical fixed policy, evaluating 81.9% fewer paths and retaining 47.2% fewer evidence items than Fixed Wide. Transfer experiments on HotpotQA, MuSiQue, and 2WikiMultiHopQA confirm 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

Original 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 at a Glance

Query-weighted Answer Correctness: 76.97 (Medical), 64.33 (Novel) on GraphRAG-Bench
+5.13 and +4.43 points over the strongest previously reported overall results
95.1 Evidence Recall and 86.1 Context Relevancy on Medical
81.9% fewer paths evaluated and 47.2% fewer evidence items retained vs. Fixed Wide
Transfers to HotpotQA, MuSiQue, and 2WikiMultiHopQA without retriever retraining