Research projects

Causal inference from unstructured data

Description: One of the central goals of causal machine learning is the accurate estimation of heterogeneous treatment effects from observational data. These estimates can guide decision-making in critical areas like personalised medicine and public policy. This research project is focused on the estimation of heterogeneous treatment effects from both structured (tabular) and unstructured (textual) data.

Research output:

• From Text to Treatment Effects: A Meta-Learning Approach to Handling Text-Based Confounding
  Henri Arno, Paloma Rabaey, and Thomas Demeester (2024)
  Workshop: Presented at the Causal Representation Learning workshop at NeurIPS 2024
  paper

• SynSUM - Synthetic Benchmark with Structured and Unstructured Medical Records
  Paloma Rabaey, Henri Arno and Thomas Demeester (2024)
  Workshop: Presented at the GenAI4Health workshop at AAAI 2025
  paper

Detection of financially distressed firms through predictive modeling

Description: In Belgium, each commercial court has one or more Chambers for Companies in Difficulty (CCDs) that detect, investigate and monitor financially distressed companies. Recognizing the inefficiencies of manually selection, the goal of this project is to assist the judges of the commercial courts in the detection of companies in financial distress through predictive modeling. To this end, we have developed and evaluated several models to detect financial distress from both the numerical accounting data and the textual disclosures in corporate filings.

Research output:

• Business Failure Prediction From Textual and Tabular Data With Sentence-Level Interpretations
  Henri Arno, Klaas Mulier, Joke Baeck and Thomas Demeester (2025)
  Journal: Annals of Operations Research (ANOR)
  Special Issue: Ensemble Learning for Operations Research and Business Analytics
  paper code

• Ontwikkeling van een AI-Model voor de Kamer voor Ondernemingen in Moeilijkheden bij de Ondernemingsrechtbank Joke Baeck, Henri Arno, Tibe Habils, Klaas Mulier and Thomas Demeester (2024)
  Journal: Tijdschrift voor Privaatrecht (TPR - Dutch)
  Special Issue: Liber Amicorum Matthias Storme
  paper

• From Numbers to Words: Multi-Modal Bankruptcy Prediction Using the ECL Dataset
  Henri Arno, Klaas Mulier, Joke Baeck and Thomas Demeester (2023)
  Workshop: Proceedings of the 6th Workshop on Financial Tech. and Natural Language Processing at IJCNLP-AACL 2023
  paper code

• Next-Year Bankruptcy Prediction from Textual Data: Benchmark and Baselines
  Henri Arno, Klaas Mulier, Joke Baeck and Thomas Demeester (2022)
  Workshop: Proceedings of the 4th Workshop on Financial Tech. and Natural Language Processing at IJCAI-ECAI 2022
  paper code