FONDUE (artiFicial intelligence fOr administrative aND bUsiness procEsses)

Kavé Salamatian




Kavé Salamatian,
Professor,
kave.salamatian@univ-smb.fr





Hervé Verjus,




Hervé Verjus,
Associate Professor,
herve.verjus@univ-smb.fr
 






 

DESCRIPTION

The FONDUE project ambition is to apply AI-based methodologies to a large set of settings relevant to business and administrative processes. Artificial intelligence (AI), like large language models (LLMs) and expert systems, is transforming business and administrative processes, impacting all three steps of business processes, namely information retrieval, regulation/rule matching, and decision-making. AI has a significant societal impact but also poses challenges, including bias, ethical concerns, and the risk of reinforcing societal inequalities. Business AI systems must integrate monitoring components from the design phase to assess ethical, social, and regulatory impacts alongside operational metrics. The FONDUE project aims to apply AI methodologies to improve all stages of business processes while addressing bias and ethical risks. It emphasizes integrating monitoring and evaluation components into AI systems to enhance transparency, explainability, and societal benefits.

The approach taken in the FONDUE project is innovative from the perspective that it uses a variety of IA tools to address issues that are frequently overlooked in favour of scientific domains like (health, (geo)physics, chemistry...). The FONDUE approach adopts a definitive multidisciplinary perspective that combines existing tools and adapts them to the business domain where human-centered data and processes are predominant; designing such AI solution dealing with business domain knowledge is challenging as this knowledge is generally informal and not well characterized. The scope of research that considers domain knowledge representation is still limited and the project ambition is to reduce, at its modest level, the gaps.
This project is supported by two major French companies.

ACTIVITIES

Research facet: an experimental generic AI pipeline is under investigation encompassing multimodal information extraction method. Such pipeline aims to cover generic blocks starting from data extraction, to data semantic assignment (i.e labelisation). The purpose is to design tunable and parameterized AI blocks; these blocks integrate AI methods suitable for large business unstructured data.
Education facet: some courses have been evolved to integrate AI methods and tools. The BPM course in Computer Science MsC Curriculum has been completely redesign: this course now deals with AI pipeline engineering for Business Processes. Such pipeline contains AI blocks (LLM, RAG) that process business textual documents and unlabeled and non-structured data. AI pipeline is introduced as a central entity in enterprise information systems.
 

LIST OF SELECTED PUBLICATIONS

  • Dong Wang, Shaoguang Yan, Yunqing Xia, Kavé Salamatian, Weiwei Deng, and Qi Zhang. 2022. Learning Supplementary NLP Features for CTR Prediction in Sponsored Search. Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD '22). ACM.
  • Dong Wang, Kavé Salamatian, Yunqing Xia, Weiwei Deng, and Qi Zhang. 2023. BERT4CTR: An Efficient Framework to Combine Pre-trained Language Model with Non-textual Features for CTR Prediction. Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD '23). ACM.
  • P. Dardouillet, K. Salamatian, H. Verjus, F. Loukil, D. Telisson and O. L. van, "Strategic Integration of Context for Fine-Tuning Topic Model Performance," 2024 IEEE 48th Annual Computers, Software, and Applications Conference (COMPSAC), Osaka, Japan, 2024, pp. 366-375.
  • Karima Boutalbi, Rafika Boutalbi, Hervé Verjus, and Kave Salamatian. 2024. Hierarchical Tensor Clustering for Multiple Graphs Representation. In Companion Proceedings of the ACM Web Conference 2024 (WWW '24). Association for Computing Machinery, New York, NY, USA, 613–616.
  • Faiza Loukil, Sarah Cadereau, Hervé Verjus, Mattéo Galfre, Kavé Salamatian, David Telisson, Quentin Kembellec, Olivier Le Van, 2024. LLM-centric pipeline for information extraction from invoices, The 2nd International Conference on Foundation and Large Language Models (FLLM2024), 26-29 November, 2024 Dubai, UAE
  • Jason Piquenot, Maxime Bérar, Pierre Héroux, Jean-Yves Ramel, Romain Raveaux, Sébastien Adam. "Finding path and cycle counting formulae in graphs with Deep Reinforcement Learning". CoRR abs/2410.01661 (2024)
  • Vasisht Duddu, Antoine Boutet, Virat Shejwalkar. Quantifying privacy leakage in graph embedding. MobiQuitous, 2020.
  • Boutet, Magnana. Anonymization by Design of Language Modeling. arXiv preprint arXiv:. (2024).
 

CHAIR PRESENTATION

Chair Presentation FONDUE

License:

Unless otherwise stated, all documents are shared under the Creative Commons BY-NC-ND 4.0 license.
You may view and share them for non-commercial purposes, without modification, and with appropriate credit to the authors.
Published on  September 12, 2025
Updated on September 12, 2025