Large Language Models for Academic Internal Auditing

This research examines the application of Artificial Intelligence in internal auditing, focusing on document management and information retrieval in academic institutions. The study proposes using Large Language Models to streamline document processing during audit preparation, addressing inefficiencies in traditional document handling methods. Through experimental evaluation of three embedding models (BGE-M3, Nomic-embed-text-v1, and CamemBERT) on a dataset of 300 academic regulatory queries, the research demonstrates BGE-M3's superior performance with an nDCG3 score of 0.90 and top-1 accuracy of 82.5%. The methodology incorporates query expansion using GPT-4 and Llama 3.1, revealing robust performance across varied query formulations. While highlighting AI's potential to transform internal auditing practices, particularly in Morocco's academic sector, the study acknowledges implementation challenges including institutional constraints and resistance to technological change. The conducted experiments and result analysis provide useful criteria that can be applied to similar information retrieval challenges in other fields and real-world applications.

@article{CHAMMAA2025,
  title     = {Large Language Models for Academic Internal Auditing},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {16},
  number    = {1},
  year      = {2025},
  publisher = {The Science and Information Organization},
  author    = {Houda CHAMMAA and Rachid ED-DAOUDI and Khadija BENAZZI},
  doi       = {10.14569/IJACSA.2025.0160166},
  url       = {https://doi.org/10.14569/IJACSA.2025.0160166}
}