DocNavRAG: Document-Structured Graph RAG with Stateful Evidence Construction for Complex Document Question Answering
2026-08-03 • Computation and Language
Computation and Language
AI summaryⓘ
The authors developed DocNavRAG to help answer complicated questions by better navigating and using information from large documents. Instead of searching from the start each time, their method treats the document like a map, moving step-by-step through related sections and pages to find useful evidence. This way, it gradually collects enough information to answer questions more accurately. They tested DocNavRAG on several challenging question sets and found it gave better answers and used more helpful context than previous methods.
Document retrievalQuestion answeringGraph navigationEvidence accumulationMulti-document QAHierarchical document structureInformation retrievalStructured retrievalAgentic retrievalContext sufficiency
Authors
Dongyang Xie, Yao Tian, Hao Zhang, Yifei Yuan, Tieyun Qian, Ming Zhong, Jiawei Jiang, Yuanyuan Zhu
Abstract
Answering complex questions over large document collections requires assembling complementary evidence across sections and documents. GraphRAG offers structured retrieval but typically uses fixed traversal, while agentic RAG operates over weakly structured interfaces. Our key insight is that agents should navigate document structure within and across documents rather than repeatedly search from scratch. We introduce DocNavRAG, which organizes document hierarchies and cross-region relations into a navigable graph, exposes graph operations for locating, navigating, expanding, and fetching, and maintains an evolving evidence state to guide retrieval until sufficient evidence is collected. Across four long- and multi-document QA benchmarks, DocNavRAG improves answer quality and context sufficiency over the strongest baseline by 7.8\% and 17.7\% on average.