RAGU: A Multi-Step GraphRAG Engine with a Compact Domain-Adapted LLM

2026-07-13

AI Summary

Existing GraphRAG systems suffer from entity noise and poor retrieval robustness due to constructing knowledge graphs via single-pass extraction. This work proposes RAGU, a novel retrieval-augmented generation engine that decouples graph construction into two stages: extraction and integration. RAGU enhances graph quality through typed information extraction, DBSCAN-based deduplication, LLM-powered summarization, and Leiden community detection. Motivated by the insight that linguistic competence and factual knowledge can be disentangled, the authors train Meno-Lite-0.1—a lightweight 7B-parameter model specialized for extraction tasks. Evaluated on GraphRAG-Bench (medical domain), RAGU achieves an evidence recall of 0.84 (surpassing baselines capped at 0.76), improves graph metrics by 12.5%, outperforms HippoRAG2 on multi-hop question answering, and operates efficiently on a single GPU.

Abstract

Graph retrieval-augmented generation (GraphRAG) enhances large language models with structured knowledge, yet existing systems construct knowledge graphs in a single extraction pass, producing noisy entities and brittle retrieval. RAGU, an open-source modular GraphRAG engine, addresses this by separating extraction from consolidation: entities and relations pass through two-stage typed extraction, DBSCAN-backed deduplication, LLM summarization, and Leiden community detection. A key insight motivates a compact extractor: the skills an in-pipeline LLM needs - comprehension, extraction, reasoning over context - are language skills that grow only weakly with model size, unlike factual world knowledge. Accordingly, we train Meno-Lite-0.1, a 7B model optimized for language skills, which outperforms Qwen2.5-32B on knowledge-graph construction (+12.5% relative harmonic mean) and matches it on English GraphRAG tasks. On GraphRAG-Bench (Medical), RAGU retrieves the most complete context at every factoid level (evidence recall up to 0.84 vs. $\leq$0.76) and overtakes HippoRAG2 on synthesis tasks; on multi-hop factoid QA, the apparent HippoRAG2 advantage is shown to be largely an answer-format artifact. RAGU is installable via $\texttt{pip install graph_ragu}$, runs on a single GPU, and is released under MIT. The source code is publicly available at https://github.com/RaguTeam/RAGU, and the Meno-Lite-0.1 model can be obtained from https://huggingface.co/bond005/meno-lite-0.1.

Problem

• GraphRAG
• knowledge graph construction
• entity extraction
• retrieval robustness
• multi-hop QA

Innovation

• GraphRAG
• knowledge graph construction
• compact LLM
• modular pipeline
• entity deduplication