@article{ART003329446},
author={Dong-Wook Shin and Nam-Mee Moon},
title={Performance Optimization Study of Hybrid RAG Engine Integrating Multi-Source Knowledge: Vector, Graph, and Ontology Approaches},
journal={Journal of The Korea Society of Computer and Information},
issn={1598-849X},
year={2026},
volume={31},
number={4},
pages={45-54}
TY - JOUR
AU - Dong-Wook Shin
AU - Nam-Mee Moon
TI - Performance Optimization Study of Hybrid RAG Engine Integrating Multi-Source Knowledge: Vector, Graph, and Ontology Approaches
JO - Journal of The Korea Society of Computer and Information
PY - 2026
VL - 31
IS - 4
PB - The Korean Society Of Computer And Information
SP - 45
EP - 54
SN - 1598-849X
AB - While RAG addresses Large Language Model(LLM) hallucinations, Vector-RAG struggles with multi-hop reasoning and logical constraints. We propose a Triple-Hybrid RAG framework combining Vector, Graph, and Ontology knowledge sources. A Dynamic Weighting Algorithm (DWA) is introduced that continuously adjusts the contribution weights of each source based on query intent signals—entity density, relation density, and constraint density—rather than relying on discrete type-based routing.
Experimental results using a synthetic university administrative dataset (1,037 unstructured text documents, 2,542 graph nodes, 6,889 edges, 5,000 gold QA) with GPT-4o-mini (temperature=0.0) showed a 19.4% improvement in F1 Score and a 34.5% gain in Exact Match(EM) score for complex queries compared to single-source RAG. A three-stage ablation study validated the contribution of each DWA component, with continuous weight adjustment yielding an additional 3.2%p Multi-hop EM improvement over type-fixed weights. Additional validation on 300 HotpotQA samples confirmed the architecture's generalizability, with F1 and EM improvements of 22.9% and 95.5%, respectively.
KW - Retrieval-Augmented Generation(RAG);Knowledge Graph;Ontology;Hybrid RAG;Multi-hop Reasoning;Dynamic Weighting Algorithm(DWA)
DO -
UR -
ER -
Dong-Wook Shin and Nam-Mee Moon. (2026). Performance Optimization Study of Hybrid RAG Engine Integrating Multi-Source Knowledge: Vector, Graph, and Ontology Approaches. Journal of The Korea Society of Computer and Information, 31(4), 45-54.
Dong-Wook Shin and Nam-Mee Moon. 2026, "Performance Optimization Study of Hybrid RAG Engine Integrating Multi-Source Knowledge: Vector, Graph, and Ontology Approaches", Journal of The Korea Society of Computer and Information, vol.31, no.4 pp.45-54.
Dong-Wook Shin, Nam-Mee Moon "Performance Optimization Study of Hybrid RAG Engine Integrating Multi-Source Knowledge: Vector, Graph, and Ontology Approaches" Journal of The Korea Society of Computer and Information 31.4 pp.45-54 (2026) : 45.
Dong-Wook Shin, Nam-Mee Moon. Performance Optimization Study of Hybrid RAG Engine Integrating Multi-Source Knowledge: Vector, Graph, and Ontology Approaches. 2026; 31(4), 45-54.
Dong-Wook Shin and Nam-Mee Moon. "Performance Optimization Study of Hybrid RAG Engine Integrating Multi-Source Knowledge: Vector, Graph, and Ontology Approaches" Journal of The Korea Society of Computer and Information 31, no.4 (2026) : 45-54.
Dong-Wook Shin; Nam-Mee Moon. Performance Optimization Study of Hybrid RAG Engine Integrating Multi-Source Knowledge: Vector, Graph, and Ontology Approaches. Journal of The Korea Society of Computer and Information, 31(4), 45-54.
Dong-Wook Shin; Nam-Mee Moon. Performance Optimization Study of Hybrid RAG Engine Integrating Multi-Source Knowledge: Vector, Graph, and Ontology Approaches. Journal of The Korea Society of Computer and Information. 2026; 31(4) 45-54.
Dong-Wook Shin, Nam-Mee Moon. Performance Optimization Study of Hybrid RAG Engine Integrating Multi-Source Knowledge: Vector, Graph, and Ontology Approaches. 2026; 31(4), 45-54.
Dong-Wook Shin and Nam-Mee Moon. "Performance Optimization Study of Hybrid RAG Engine Integrating Multi-Source Knowledge: Vector, Graph, and Ontology Approaches" Journal of The Korea Society of Computer and Information 31, no.4 (2026) : 45-54.
