«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

期数:

2025年07期

页码:

117-124

栏目:

人工智能

出版日期:

2025-07-10

文章信息/Info

Title:

Research on Ontology Evolution Mechanism Driven by Large Language Models

文章编号:

1673-629X(2025)07-0117-08

作者:

廖光忠1; 2; 方锤1; 2; 陆晨阳1; 2; 顾进广1; 2

1. 武汉科技大学 计算机科学与技术学院,湖北 武汉 430065;

2. 湖北省智能信息处理与实时工业系统重点实验室,湖北 武汉 430065

Author(s):

LIAO Guang-zhong1; 2; FANG Chui1; 2; LU Chen-yang1; 2; GU Jin-guang1; 2

1. School of Computer Science and Technology,Wuhan University of Science and Technology,Wuhan 430065,China;

2. Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial,Wuhan 430065,China

关键词:

本体; 本体演化; 提示工程; 范式; 大型语言模型

Keywords:

ontology; ontology evolution; prompt engineering; paradigm; large language models

分类号:

TP311.1

DOI:

10.20165/j.cnki.ISSN1673-629X.2025.0047

摘要:

在本体演化过程中,尽管领域专家的参与至关重要,但也同时存在着知识局限、成本高昂、效率低下等问题。 为此,该文提出了一种利用大语言模型来指导本体演化的方法,旨在降低本体演化过程对领域专家的依赖。 方案共设计六种提示范式:二元决策完型范式、二元决策前缀范式、多元决策完型范式、多元决策前缀范式、整体输入范式、分步输入范式。 通过四种方式(二元决策、多元决策、完整输入和分步输入)将待删除的实体三元组等信息输入到大模型中,根据模型输出结果重新构建本体,比较本体演化前后的结构相似率来衡量模型效果。 研究使用了 ChatGPT-V3. 5 和 Llama-7B 两种模型,并对民航客服本体(CACSO)、天气预测本体(SENSOR)和中文糖尿病本体(CDMO)三个领域的数据集进行了零样本实验。 实验结果表明,与传统的专家驱动方法相比,基于大语言模型的本体演化方法显著提升了效率,降低了对专家资源的依赖,验证了大语言模型在本体演化中的可行性,为本体演化领域提供了一种创新的智能化决策方法。

Abstract:

In the ontology evolution process,although the involvement of domain experts is crucial,it also presents challenges such as knowledge limitations,high costs,and low efficiency. To address these issues,we propose a approach that utilizes large language models to guide ontology evolution, aiming to reduce the dependency on domain experts. The method designs a total of six prompting paradigms:binary decision cloze paradigm,binary decision prefix paradigm,multivariate decision cloze paradigm,multivariate decision prefix paradigm,disposable input paradigm and step-by-step input paradigm. Information such as entity triples to be deleted is input into the large model through four methods (binary decision,multivariate decision,disposable input,and step-by-step input). The ontology is then reconstructed based on the model outputs,and the structural similarity between pre- and post-evolution ontologies is compared to e-valuate the model’s performance. We employ two models,ChatGPT- V3. 5 and Llama - 7B,and conduct zero - shot experiments on datasets from three domains:Civil Aviation Customer Service Ontology (CACSO),Weather Forecast Ontology (SENSOR),and Chinese Diabetes Mellitus Ontology (CDMO). The experimental results demonstrate that compared to traditional expert - driven methods,the large language model-based approach significantly enhances efficiency,reduces reliance on expert resources,and validates the feasibility of large language models in ontology evolution,providing an innovative and intelligent decision-making method for the field.

常用功能

更新日期/Last Update: 2025-07-10