Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Democratizing GraphRAG: Linear, CPU-Only Graph Retrieval for Multi-Hop QA

Qizhi Wang

Abstract

GraphRAG systems improve multi-hop retrieval by modeling structure, but many approaches rely on expensive LLM-based graph construction and GPU-heavy inference. We present SPRIG (Seeded Propagation for Retrieval In Graphs), a CPU-only, linear-time, token-free GraphRAG pipeline that replaces LLM graph building with lightweight NER-driven co-occurrence graphs and uses Personalized PageRank (PPR) for 28% with negligible Recall@10 changes. The results characterize when CPU-friendly graph retrieval helps multi-hop recall and when strong lexical hybrids (RRF) are sufficient, outlining a realistic path to democratizing GraphRAG without token costs or GPU requirements.

Democratizing GraphRAG: Linear, CPU-Only Graph Retrieval for Multi-Hop QA

Abstract

GraphRAG systems improve multi-hop retrieval by modeling structure, but many approaches rely on expensive LLM-based graph construction and GPU-heavy inference. We present SPRIG (Seeded Propagation for Retrieval In Graphs), a CPU-only, linear-time, token-free GraphRAG pipeline that replaces LLM graph building with lightweight NER-driven co-occurrence graphs and uses Personalized PageRank (PPR) for 28% with negligible Recall@10 changes. The results characterize when CPU-friendly graph retrieval helps multi-hop recall and when strong lexical hybrids (RRF) are sufficient, outlining a realistic path to democratizing GraphRAG without token costs or GPU requirements.
Paper Structure (29 sections, 3 equations, 6 figures, 26 tables)

This paper contains 29 sections, 3 equations, 6 figures, 26 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Pipeline Overview
  5. Graph Construction (Linear-Time)
  6. Retrieval via PPR
  7. Complexity and Memory
  8. Experimental Setup
  9. Datasets
  10. Evaluation Metrics
  11. Baselines
  12. Graph and PPR Settings
  13. QTime definition.
  14. Hyperparameter Tuning Protocol
  15. Environment
  16. ...and 14 more sections

Figures (6)

  • Figure 1: Efficiency curves for index and query time.
  • Figure A.1: Seed size ablation (Recall@10).
  • Figure A.2: Ablation over PPR $\alpha$ and max_iter (Recall@10).
  • Figure A.3: Seed weighting ablation (best Recall@10).
  • Figure A.4: Hub penalty ablation (best Recall@10).
  • ...and 1 more figures