Abstract:
Ontology Engineering (OE) typically begins with ontology requirements elicitation, in which ontology engineers define the scope of concepts and relations an ontology must cover to adequately serve its intended application domain. Large Language Models (LLMs) can generate a large set of Competency Question (CQ) candidates to support this process; however, a target ontology scope is often multi-dimensional, ill-defined, and cannot be fully anticipated in advance, and existing tools provide little support for ontology engineers to subsequently explore and generate new CQ candidates (divergent thinking), evaluate, refine, and eliminate existing ones (convergent thinking), and thus progressively define a well-scoped ontology. We argue that interaction designs from LLM-based systems in arts and creativity domains, where divergent and convergent thinking have been extensively studied, are transferable to ontology requirements elicitation. To identify these designs, we conducted a Systematic Literature Review (SLR) of 50 papers, identifying 7 Interaction Techniques (ITs) and 14 User Interfaces (UIs), each justified with respect to how it supports divergent thinking, convergent thinking, or both. To explore the transferability and applicability of the identified ITs and UIs to LLM-based ontology scoping, we conducted a design thinking workshop (N=7) that produced a conceptual interaction model, a system prototype called OntoScope implementing that model, and a use case demonstration showing how OntoScope can potentially support an OE expert in scoping a university ontology. The identified ITs and UIs can serve as a reference for tool developers working on ontology requirements elicitation, broader OE tasks that require human reasoning and auditing over LLM-generated content, or designing user-friendly OE tools for domain experts and end users without prior OE expertise.
