Automated and LLM-Assisted Conceptual Modeling from Semi-structured Data

This chapter delves into the complexities of conceptual modeling, highlighting the manual nature of traditional methods and the challenges posed by semi-structured data. It introduces an automated, LLM-assisted approach to generate Entity-Relationship Diagrams (ERDs), focusing on overcoming issues like ambiguous naming and incomplete identifiers. The method employs an iterative cycle, incorporating expert input to refine ERDs and improve contextual accuracy. Evaluation metrics such as FieldCoverage, 3NF Compliance, and Referential Integrity are used to assess the quality of generated ERDs. The chapter also presents a tool designed to support this method, detailing its architecture and deployment. Experimental results on two datasets demonstrate the effectiveness of the approach, showing how varying temperature settings impact model performance. The findings underscore the importance of expert refinement in achieving accurate and domain-aligned conceptual models, despite challenges in achieving full 3NF compliance and capturing business rules. This chapter offers valuable insights into leveraging LLMs for data engineering tasks, providing a practical framework for professionals to enhance their conceptual modeling processes.