LeanRAG: Knowledge-Graph-Based Generation with Semantic Aggregation and Hierarchical Retrieval (2508.10391v2)

Abstract: Retrieval-Augmented Generation (RAG) plays a crucial role in grounding LLMs by leveraging external knowledge, whereas the effectiveness is often compromised by the retrieval of contextually flawed or incomplete information. To address this, knowledge graph-based RAG methods have evolved towards hierarchical structures, organizing knowledge into multi-level summaries. However, these approaches still suffer from two critical, unaddressed challenges: high-level conceptual summaries exist as disconnected ``semantic islands'', lacking the explicit relations needed for cross-community reasoning; and the retrieval process itself remains structurally unaware, often degenerating into an inefficient flat search that fails to exploit the graph's rich topology. To overcome these limitations, we introduce LeanRAG, a framework that features a deeply collaborative design combining knowledge aggregation and retrieval strategies. LeanRAG first employs a novel semantic aggregation algorithm that forms entity clusters and constructs new explicit relations among aggregation-level summaries, creating a fully navigable semantic network. Then, a bottom-up, structure-guided retrieval strategy anchors queries to the most relevant fine-grained entities and then systematically traverses the graph's semantic pathways to gather concise yet contextually comprehensive evidence sets. The LeanRAG can mitigate the substantial overhead associated with path retrieval on graphs and minimizes redundant information retrieval. Extensive experiments on four challenging QA benchmarks with different domains demonstrate that LeanRAG significantly outperforming existing methods in response quality while reducing 46\% retrieval redundancy. Code is available at: https://github.com/RaZzzyz/LeanRAG

• The paper introduces a framework using hierarchical knowledge graph aggregation to resolve contextual gaps and redundant retrieval in RAG systems.
• It employs a two-step process with recursive semantic clustering via Gaussian Mixture Models and lean LLM-driven abstraction to form unified entity clusters.
• Experimental results show a 46% reduction in redundancy and enhanced response quality across diverse QA benchmarks in domains like CS, Legal, and Agriculture.

LeanRAG: Knowledge-Graph-Based Generation with Semantic Aggregation and Hierarchical Retrieval

Introduction

The paper "LeanRAG: Knowledge-Graph-Based Generation with Semantic Aggregation and Hierarchical Retrieval" introduces a framework that advances the field of Retrieval-Augmented Generation (RAG). Traditional RAG methods face challenges in retrieving and using contextually flawed or incomplete information from large texts. Leveraging knowledge graph-based methodologies, LeanRAG aims to construct a hierarchically organized semantic network from existing data, resolving issues associated with disjoint "semantic islands" and unstructured, inefficient retrieval processes.

Figure 1: Comparison of typical LLM retrieval-augmented generation frameworks.

Methodology

Hierarchical Knowledge Graph Aggregation

LeanRAG employs a two-step process. Initially, it utilizes a semantic aggregation algorithm to create semantically similar entity clusters, enhancing the overall structure and richness of the knowledge graph. The method involves:

• Recursive Semantic Clustering: Entities are clustered using a Gaussian Mixture Model (GMM). Semantic embeddings of entity texts are used to partition the entities, forming clusters that serve as the basis for further abstraction.
• Aggregated Entity and Relation Generation: A lean approach driven by LLMs facilitates the creation of new, abstract entities and their interrelations. This step is crucial in transforming clustering outcomes into a structured, unified network.

Structured Retrieval Strategy

LeanRAG's retrieval process is structure-aware, focusing on efficient, precise retrieval paths:

• Entity Anchoring: Initial retrieval identifies top relevant entities, anchoring subsequent graph-based traversal.
• LCA-Based Path Exploration: Leveraging the Lowest Common Ancestor (LCA) retrieval concept, path finding is conducted across hierarchical levels, enabling a comprehensive yet concise retrieval of relevant information.

  Figure 2: Overview of the LeanRAG framework.

Experimental Evaluation

LeanRAG was evaluated across four QA benchmarks encompassing various domains such as CS, Legal, and Agriculture. The experimental results demonstrated its superiority in response quality, also reflecting a 46% reduction in information redundancy when compared to baselines. LeanRAG consistently outperformed other contemporary methods, such as GraphRAG and HiRAG, across metrics like Comprehensiveness, Empowerment, and Diversity.

Figure 3: Comparison in retrieval tokens across four datasets.

Implications and Future Work

The practical implications of LeanRAG involve its potential for significantly enhancing LLM-based applications across diverse areas by reducing redundant retrieval and improving contextual accuracy in generated texts. Future work might explore more fine-tuned hyperparameter adjustments and expansions to other domains or languages, assessing LeanRAG's adaptability and robustness further.

Conclusion

LeanRAG represents a notable advance in RAG frameworks by amalgamating hierarchical abstraction with semantic precision in entity retrieval. This integration across multiple levels of knowledge graph abstraction is pivotal in addressing previous limitations, positioning LeanRAG as a vital tool in the landscape of knowledge-intensive LLM generation.