An Exploratory Study of Conceptual Modeling for Data Warehousing using LLMs

With a growing number of applications of AI in various fields, we discovered a transformative potential to improve efficiency and aid humans in different tasks. This thesis is aimed to research how Large Language Models (LLMs) can be used in conceptual modeling of data warehouses using the Dimensional Fact Model (DFM) and facilitate the process by actively leveraging this new technology. First, theoretical foundations of LLMs and conceptual models are explored in detail, describing the underlying architectures, capabilities and conducting a comprehensive overview. Then, using prompt engineering tailored with different inputs to get the resulting outcomes, we analyze these outputs by using a binary classification confusion matrix and evaluate them with measures such as precision, recall and F1 score. Results indicate that LLMs are capable of generating fact schema, although performing best when given examples or a desired output format. For all other cases, models may have varying performance. Also, the research highlights the limitations and challenges for completing complex logical reasoning tasks.