Extraction d'informations dans des rapports d'incidents du réseau électrique à l'aide de LLMs open source

Effective management of an organization’s knowledge requires processing numerous textual documents. In the context of French power grid incident reports, this task is particularly challenging due to the specificity of the texts and the complexity of the information to be extracted. This study explores the use of large language models (LLMs) to automatically extract relevant information with minimal annotation effort. We experiment with Information Extraction (IE) using various strategies and evaluate each of them quantitatively. Our findings provide insights into the most efficient methods for leveraging LLMs for IE and offer practical guidance to apply them in real-world scenarios, such as the management of power grid incidents.