An ETL Method for Data Warehouse Design Driven by Ontology and LLM
- Lecture Notes in Networks and Systems, vol 2117. Springer, Cham. , 2117 : 338-346
Résumé
The processes involved in feeding a warehouse require such a high level of expertise that they are generally entrusted to an IT specialist (data administrator). This requires, first, that decision-makers express their needs to the IT specialist and, second, that the data warehouse developed by this IT specialist be validated. In the era of open decision-making information (self-service BI), decision-makers increasingly want to integrate and analyze data themselves without the help of experts. Added to this is the diversity of data, its speed of evolution, and its ever-increasing volume, which hinder traditional ETL approaches because they can no longer meet new requirements. We therefore offer an assistance process that allows decision-makers to integrate data themselves (Extract-Transform-Load: ETL) into a data warehouse using Large Language Models (LLMs) and ontology, without the need for an IT specialist. This work focuses on the needs(in natural language) of decision-makers by mapping them to a domain ontology and generating the metadata needed to guide data extraction and consolidation.
Mots-clés
Extract-Transform-Load (ETL); Large Language Models (LLMs) ; Ontology ; data warehousing ; Natural language; Requirements