KG2QA: Knowledge Graph-enhanced Retrieval-augmented Generation for Communication Standards Question Answering
Paper: https://arxiv.org/abs/2506.07037
Zhongze Luo1, Weixuan Wan4, Tianya Zhang1, Dan Wang4, Xiaoying Tang1,2,3*
1 School of Science and Engineering, The Chinese University of Hong Kong, Shenzhen, China
2 The Shenzhen Future Network of Intelligence Institute (FNii-Shenzhen), China
3 The Guangdong Provincial Key Laboratory of Future Networks of Intelligence, China
4 School of Microelectronics, Xi’an Jiaotong University, China
*Corresponding author: Xiaoying Tang
The rapid evolution of communication technologies has led to an explosion of standards, rendering traditional expert-dependent consultation methods inefficient and slow. To address this challenge, we propose KG2QA, an intelligent question-answering system for communication standards that integrates fine-tuned large language models (LLMs) with a domain-specific knowledge graph (KG) via a Retrieval-Augmented Generation (RAG) framework. We construct a high-quality dataset of 6,587 QA pairs from ITU-T recommendations and fine-tune Qwen2.5-7B-Instruct, achieving significant performance gains: BLEU-4 increases from 18.86 to 66.90, outperforming both the base model and Llama-3-8B-Instruct. A structured KG containing 13,906 entities and 13,524 relations is built using LLM-assisted triple extraction based on a custom ontology. In our RAG pipeline, the fine-tuned LLM first retrieves relevant knowledge from the KG, enabling more accurate and factually grounded responses. Evaluated by DeepSeek-V3 as a judge, the KG-enhanced system improves performance across five dimensions, with an average score increase of 2.26%, demonstrating superior factual accuracy and relevance. Integrated with Web platform and API, KG2QA delivers an efficient and interactive user experience. Our code and data have been open-sourced.
@inproceedings{luo2026kg2qa,
title={KG2QA: Knowledge Graph-enhanced Retrieval-augmented Generation for Communication Standards Question Answering},
author={Zhongze Luo and Weixuan Wan and Tianya Zhang and Dan Wang and Xiaoying Tang}
booktitle={ICASSP 2026-2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
year={2026},
organization={IEEE}
}
This dataset can be used for fine-tuning large language models in the field of communication standards. The data is sourced from ITU-T Recommendations and the language is English. You can click here to visit the ITU-T official website and view the source file.
The dataset is divided into the training set and the test set. The training set is composed of data of the three categories of G, V, X, all of which are saved in JSON, and the test set is test.json.
The dataset was obtained by QA extraction of the source file PDF using the Deepseek V3 API, and the data volume is as follows.
288 PDF source files are stored in the G, V, X folders corresponding to their category sources.
| Category | Quantity of QA |
|---|---|
| G.json | 898 |
| V.json | 1,101 |
| X.json | 3,779 |
| test.json | 809 |
| All Categories | 6,587 |
The data is derived from ITU-T Recommendations and is taken from three series of recommendations. A total of 288 PDF source files were selected for QA construction in the three series. The specific category sources are as follows.
G series： Transmission systems and media, digital systems and networks (A total of 34 PDF source files)
G.600-G.699：Transmission media and optical systems characteristics
G.650-G.659：Optical fibre cables
G.660-G.679：Characteristics of optical components and subsystems
G.680-G.699： Characteristics of optical systems
V series：Data communication over the telephone network (A total of 77 PDF source files)
V.1-V.9：General
V.10-V.34：Interfaces and voiceband modems
V.35-V.39：Wideband modems
V.40-V.49：Error control
V.50-V.59：Transmission quality and maintenance
V.60-V.99： Simultaneous transmission of data and other signals
V.100-V.199：Interworking with other networks
V.200-V.249：Interface layer specifications for data communication
V.250-V.299：Control procedures
V.300-V.399：Modems on digital circuits
X series：Data networks, open system communications and security (A total of 177 PDF source files)
X.1-X.199：Public data networks
X.1-X.19：Services and facilities
X.20-X.49：Interfaces
X.50-X.89：Transmission, signalling and switching
X.90-X.149：Network aspects
X.150-X.179：Maintenance
X.180-X.199：Administrative arrangements
X.300-X.399：Interworking between networks
X.300-X.349：General
X.350-X.369：Satellite data transmission systems
X.370-X.379：IP-based networks
X.1000-X.1099：Information and network security
X.1000-X.1011：Security orchestration and service access
X.1030-X.1049： Network security
X.1050-X.1079：Security management
X.1080-X.1099：Telebiometrics
X.1700-X.1729：Quantum communication
X.1702-X.1704：Quantum random number generator
X.1705-X.1749：Quantum Key Distribution Network (QKDN)
X.1750-X.1799：Data security
X.1750-X.1759：Big Data Security
X.1770-X.1799：Data protection (II)
Fine-tuning Experimental Setting
Fine-tuned Model Evaluation Code (DeepSeek example)
Ontology are saved in the kg folder. After opening Protege.exe, click File-Open and select on.rdf to open it
- Node and relationship data are saved in the
importfolder in CSV UTF-8 format. Save this folder in the local Neo4j path, and then ForPKG can be stored and displayed.
D:\neo4j-community-5.18.1\import
- Restart neo4j
On the computer, type
cmdto enter the command line. Go toneo4j-community-4.3.18\bin, typeneo4j restartto restart neo4j. In the browser, typelocalhost:7474/browser/to enter neo4j.
cd D:\neo4j-community-5.18.1\bin
neo4j restart
- Click the database icon and click
:dbsof DBMS
1. Import a specific entity type
LOAD CSV WITH HEADERS FROM 'file:///file_name.csv' AS line # file_name represents a specific entity type, and file_name.csv is the CSV file for that entity type
MERGE (:file_name { ID: line.ID, name: line.name, LABEL: line.LABEL })
2. Delete a specific entity type
MATCH (r:`file_name`) DETACH DELETE r // file_name represents a specific entity type
3. Create all relationships
LOAD CSV WITH HEADERS FROM 'file:///roles.csv' AS row // roles.csv is the CSV file containing all relationships
MATCH (fromNode {ID: row.from}), (toNode {ID: row.to})
CALL apoc.create.relationship(fromNode, row.relation, {}, toNode) YIELD rel
RETURN rel
4. Delete a specific relationship
MATCH ()-[r:duty]-() DETACH DELETE r
5. Delete all relationships
MATCH ()-[r]-() DETACH DELETE r
6. Display all entities and relationships
MATCH (n) RETURN n
ID,name,LABEL
from,to,relation
Ollama Model Remote Access to Knowledge Graph
Quantitative Evaluation of Answer Quality using an LLM-based Judge
Knowledge Graph QA Web Service Platform
Knowledge Graph QA API Service Platform
This work is supported in part by the Guangdong Basic and Applied Basic Research Foundation under Grant No. 2025A1515012968, Shenzhen Science and Technology Program under Grant No. JCYJ20240813113502004, National Natural Science Foundation of China under Grant No. 62001412, in part by the funding from Shenzhen Institute of Artificial Intelligence and Robotics for Society, in part by Shenzhen Stability Science Program 2023, and in part by the Guangdong Provincial Key Laboratory of Future Networks of Intelligence (Grant No. 2022B1212010001).