Citation

Chong, You Li and Lee, Chin Poo and Lim, Kian Ming (2025) Distilcyphergpt: enhancing large language models for knowledge graph question answering in cypher through knowledge distillation. Data Mining and Knowledge Discovery, 39 (6). ISSN 1384-5810

Text s10618-025-01157-9.pdf - Published Version Restricted to Repository staff only Download (3MB)

Abstract

Knowledge Graph Question Answering (KGQA) systems allow users to interact with knowledge graphs using natural language queries, which are translated into structured database queries like Cypher. Existing KGQA approaches often rely on large language models, leading to high computational costs and slower inference times that impede real-time applications. To address these challenges, DistilCypherGPT is introduced as an efficient KGQA framework employing knowledge distillation in a teacher-student architecture, optimized for Cypher query generation on academic knowledge graphs. DistilCypherGPT significantly reduces computational demands, enabling deployment in resource-constrained environments while retaining high accuracy. Experimental results show that DistilCypherGPT maintains 99.51% accuracy, achieving a 23% reduction in model size and a 30% improvement in inference speed compared to the baseline. These findings corroborate DistilCypherGPT’s potential as a scalable, high-performance solution for KGQA, advancing efficient, real-time query translation with minimal computational overhead.

Item Type:	Article
Uncontrolled Keywords:	Knowledge graph, Question answering, Language model, Knowledge distillation, Generative pretrained transformers
Subjects:	H Social Sciences > HD Industries. Land use. Labor > HD28-70 Management. Industrial Management > HD30.2 Electronic data processing. Information technology. Including artificial intelligence and knowledge management
Divisions:	Faculty of Information Science and Technology (FIST)
Depositing User:	Ms Suzilawati Abu Samah
Date Deposited:	30 Sep 2025 02:48
Last Modified:	05 Oct 2025 02:54
URII:	http://shdl.mmu.edu.my/id/eprint/14545

Downloads

Edit (login required)