Automating Computation Independent Model Elicitation in MDA using Task-Oriented Dialogue with In-Context Learning

Abstract

The Computation Independent Model (CIM) is a cornerstone of the Object Management Group's (OMG) Model-Driven Architecture (MDA), capturing business requirements and domain knowledge independent of specific technologies. However, the elicitation of CIM requirements is often a manual, time-consuming, and error-prone process, susceptible to ambiguities inherent in natural language. Traditional Natural Language Understanding (NLU) approaches, particularly intent-based systems, exhibit limitations in scalability, contextual understanding, and handling the nuanced, evolving nature of complex requirements. This study proposes a novel approach that integrates Task-Oriented Dialogue (TOD) systems with the In-Context Learning (ICL) capabilities of Large Language Models (LLMs) to automate and enhance CIM requirements elicitation. The proposed framework features a conversational agent that guides stakeholders through structured dialogue flows, translating their natural language inputs into a formal CIM-Domain Specific Language (CIM-DSL). These DSL commands are then transformed into CIM artifacts, such as Business Process Model and Notation (BPMN) diagrams and Unified Modeling Language (UML) use cases. The approach emphasizes quality assurance through interactive validation, consistency checks, and strategies to mitigate LLM limitations. We anticipate this method will significantly improve the accuracy, completeness, and efficiency of CIM construction, thereby strengthening the foundation of the MDA lifecycle.