Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep GraphRAG: A Balanced Approach to Hierarchical Retrieval and Adaptive Integration

Yuejie Li, Ke Yang, Tao Wang, Bolin Chen, Bowen Li, Chengjun Mao

TL;DR

Deep GraphRAG addresses the global-local retrieval trade-off in graph-based retrieval-augmented generation by introducing a hierarchical global-to-local framework with a three-stage retrieval and a beam-search-inspired re-ranking mechanism. It combines a structurally-aware knowledge graph with a three-level community hierarchy and context-aware representations, enabling efficient top-down search and fine-grained entity retrieval. The paper also introduces DW-GRPO, a dynamic weighting scheme for multi-objective reinforcement learning, which uses adaptive reward coefficients to balance relevance, faithfulness, and conciseness, allowing compact 1.5B LLMs to approach 70B-level performance in knowledge integration. Empirical results on Natural Questions and HotpotQA show state-of-the-art EM-Total and significant latency reductions, validating the approach for scalable, accurate retrieval in large-scale hierarchical graphs.

Abstract

Graph-based Retrieval-Augmented Generation (GraphRAG) frameworks face a trade-off between the comprehensiveness of global search and the efficiency of local search. Existing methods are often challenged by navigating large-scale hierarchical graphs, optimizing retrieval paths, and balancing exploration-exploitation dynamics, frequently lacking robust multi-stage re-ranking. To overcome these deficits, we propose Deep GraphRAG, a framework designed for a balanced approach to hierarchical retrieval and adaptive integration. It introduces a hierarchical global-to-local retrieval strategy that integrates macroscopic inter-community and microscopic intra-community contextual relations. This strategy employs a three-stage process: (1) inter-community filtering, which prunes the search space using local context; (2) community-level refinement, which prioritizes relevant subgraphs via entity-interaction analysis; and (3) entity-level fine-grained search within target communities. A beam search-optimized dynamic re-ranking module guides this process, continuously filtering candidates to balance efficiency and global comprehensiveness. Deep GraphRAG also features a Knowledge Integration Module leveraging a compact LLM, trained with Dynamic Weighting Reward GRPO (DW-GRPO). This novel reinforcement learning approach dynamically adjusts reward weights to balance three key objectives: relevance, faithfulness, and conciseness. This training enables compact models (1.5B) to approach the performance of large models (70B) in the integration task. Evaluations on Natural Questions and HotpotQA demonstrate that Deep GraphRAG significantly outperforms baseline graph retrieval methods in both accuracy and efficiency.

Deep GraphRAG: A Balanced Approach to Hierarchical Retrieval and Adaptive Integration

TL;DR

Deep GraphRAG addresses the global-local retrieval trade-off in graph-based retrieval-augmented generation by introducing a hierarchical global-to-local framework with a three-stage retrieval and a beam-search-inspired re-ranking mechanism. It combines a structurally-aware knowledge graph with a three-level community hierarchy and context-aware representations, enabling efficient top-down search and fine-grained entity retrieval. The paper also introduces DW-GRPO, a dynamic weighting scheme for multi-objective reinforcement learning, which uses adaptive reward coefficients to balance relevance, faithfulness, and conciseness, allowing compact 1.5B LLMs to approach 70B-level performance in knowledge integration. Empirical results on Natural Questions and HotpotQA show state-of-the-art EM-Total and significant latency reductions, validating the approach for scalable, accurate retrieval in large-scale hierarchical graphs.

Abstract

Graph-based Retrieval-Augmented Generation (GraphRAG) frameworks face a trade-off between the comprehensiveness of global search and the efficiency of local search. Existing methods are often challenged by navigating large-scale hierarchical graphs, optimizing retrieval paths, and balancing exploration-exploitation dynamics, frequently lacking robust multi-stage re-ranking. To overcome these deficits, we propose Deep GraphRAG, a framework designed for a balanced approach to hierarchical retrieval and adaptive integration. It introduces a hierarchical global-to-local retrieval strategy that integrates macroscopic inter-community and microscopic intra-community contextual relations. This strategy employs a three-stage process: (1) inter-community filtering, which prunes the search space using local context; (2) community-level refinement, which prioritizes relevant subgraphs via entity-interaction analysis; and (3) entity-level fine-grained search within target communities. A beam search-optimized dynamic re-ranking module guides this process, continuously filtering candidates to balance efficiency and global comprehensiveness. Deep GraphRAG also features a Knowledge Integration Module leveraging a compact LLM, trained with Dynamic Weighting Reward GRPO (DW-GRPO). This novel reinforcement learning approach dynamically adjusts reward weights to balance three key objectives: relevance, faithfulness, and conciseness. This training enables compact models (1.5B) to approach the performance of large models (70B) in the integration task. Evaluations on Natural Questions and HotpotQA demonstrate that Deep GraphRAG significantly outperforms baseline graph retrieval methods in both accuracy and efficiency.
Paper Structure (15 sections, 5 equations, 3 figures, 1 table, 1 algorithm)

This paper contains 15 sections, 5 equations, 3 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Methods
  3. Deep GraphRAG
  4. Graph Construction and Hierarchy
  5. Retrieval Process
  6. Dynamic Weighting Reward GRPO (DW-GRPO)
  7. DW-GRPO in Deep GraphRAG
  8. Dynamic Weighting Reward-GRPO
  9. Experimental Setup
  10. Datasets
  11. Question Categorization and Baselines
  12. Results
  13. Performance and Efficiency Analysis
  14. Performance of DW-GRPO
  15. Conclusion

Figures (3)

  • Figure 1: Deep GraphRAG framework overview. The retrieval module uses Graph Beam Search and Context Aware Ranking on a hierarchical Knowledge Graph. The knowledge integration module employs Dynamic Weighting Reward GRPO (DW GRPO).
  • Figure 2: Comparison of system processing time (latency) among Local Search (LS), DRIFT Search (DS), and Deep GraphRAG methods on the NQ dataset.
  • Figure 3: Comparison of smoothed reward learning curves between GRPO and DW-GRPO on the test dataset.