GraphSearch: An Agentic Deep Searching Workflow for Graph Retrieval-Augmented Generation

2025-09-26

AI Summary

Existing GraphRAG methods suffer from two key bottlenecks: shallow evidence retrieval and inefficient utilization of structured graph data, severely limiting their reasoning capability for complex multi-hop queries. To address these limitations, we propose Dual-Channel GraphRAG—a retrieval-augmented generation framework featuring dual parallel retrieval channels: a semantic channel (based on text fragment matching) and a relational channel (driven by graph-structure traversal). These channels jointly orchestrate iterative, agent-style deep search to enable joint reasoning over knowledge graphs and textual chunks. The framework adopts a modular, workflow-driven architecture supporting iterative multi-hop inference and progressive evidence refinement. Evaluated on six mainstream multi-hop RAG benchmarks, our approach achieves significant improvements in answer accuracy and generation quality. Notably, it provides the first systematic empirical validation that the semantic–relational dual-channel mechanism substantially enhances factual reasoning capabilities in complex question answering.

Technology Category

• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Graph Retrieval-Augmented Generation (GraphRAG) enhances factual reasoning in LLMs by structurally modeling knowledge through graph-based representations. However, existing GraphRAG approaches face two core limitations: shallow retrieval that fails to surface all critical evidence, and inefficient utilization of pre-constructed structural graph data, which hinders effective reasoning from complex queries. To address these challenges, we propose extsc{GraphSearch}, a novel agentic deep searching workflow with dual-channel retrieval for GraphRAG. extsc{GraphSearch} organizes the retrieval process into a modular framework comprising six modules, enabling multi-turn interactions and iterative reasoning. Furthermore, extsc{GraphSearch} adopts a dual-channel retrieval strategy that issues semantic queries over chunk-based text data and relational queries over structural graph data, enabling comprehensive utilization of both modalities and their complementary strengths. Experimental results across six multi-hop RAG benchmarks demonstrate that extsc{GraphSearch} consistently improves answer accuracy and generation quality over the traditional strategy, confirming extsc{GraphSearch} as a promising direction for advancing graph retrieval-augmented generation.

Problem

• Addresses shallow retrieval limitations in GraphRAG systems
• Improves utilization of pre-constructed structural graph data
• Enhances reasoning capabilities for complex multi-hop queries

Innovation

• Agentic deep searching workflow for GraphRAG
• Dual-channel retrieval over text and graph data
• Modular framework enabling multi-turn iterative reasoning