OntoChat: a Framework for Conversational Ontology Engineering using Language Models

大型项目中的本体工程（OE）面临许多挑战，这些挑战源于各方利益相关者、领域专家及其与本体设计师之间的复杂互动和异构背景。这种多方互动经常会产生系统性的模糊和偏见，这直接影响设计、评估，并可能危及目标重用。同时，当前OE方法学强烈依赖于手动活动（例如，采访、讨论页面）。在收集了关于最关键OE活动的证据之后，我们介绍了OntoChat，这是一个支持需求获取、分析和测试的对话本体工程框架。通过与对话代理交互，用户可以引导用户故事的创建和能力问题的提取，同时接收计算支持以分析整体需求并测试生成本体的早期版本。我们通过复制音乐元本体的工程，并从用户那里收集每个组件有效性的初步指标来评估OntoChat。我们在https://github.com/King-s-Knowledge-Graph-Lab/OntoChat上发布了所有代码。

Ontology engineering in large projects is challenging due to the heterogeneous backgrounds of stakeholders and the reliance on manual activities. The paper aims to introduce a conversational framework to support requirement elicitation, analysis, and testing.

The key idea of the paper is OntoChat, a conversational agent that interacts with users to create user stories, extract competency questions, and analyze requirements. OntoChat also supports early testing of resulting ontologies. This approach improves upon current OE methodologies that rely heavily on manual activities.

The paper evaluates OntoChat by replicating the engineering of the Music Meta Ontology and collecting preliminary metrics on the effectiveness of each component from users. All code is released on GitHub. The conversational framework provides computational support to analyze requirements and test early versions of ontologies.

Related work in this field includes previous OE methodologies that rely on manual activities, as well as other conversational agents for ontology engineering such as OntoGenie and OntoTrix.