publications | Mathieu Laï-king

2025

PhD Thesis

Qualité des articles de recherche et modèles de langue neuronaux : applications au domaine biomédical

Mathieu Laï-king

Apr 2025

@phdthesis{laiking:tel-05079724,
  title = {{Qualité des articles de recherche et modèles de langue neuronaux : applications au domaine biomédical}},
  author = {Laï-king, Mathieu},
  url = {https://theses.hal.science/tel-05079724},
  number = {2025UPASG031},
  school = {{Université Paris-Saclay}},
  year = {2025},
  month = apr,
  keywords = {Language model ; Data quality ; Scientific articles ; Biomedical domain ; Modèles de langue ; Qualité des données ; Articles scientifiques ; Domaine biomédical},
  type = {Theses},
  hal_id = {tel-05079724},
  hal_version = {v1},
  bibtex_show = true
}

TAL 65-2
Évaluation de la qualité de rapport des essais cliniques avec des larges modèles de langue [Evaluation of Clinical Trials Reporting Quality using Large Language Models]

Mathieu Laï-king, and Patrick Paroubek

In Traitement Automatique des Langues, Volume 65, Numéro 2 : Numéro spécial sur le Traitement automatique de documents scientifiques [Scientific Documents Processing], Apr 2025

Abs Bib PDF Code

Reporting quality is an important topic in clinical trial research articles, as it can have an impact on the clinical decisions made. In this article, we test the ability of large language models to assess the reporting quality of this type of article using the Consolidated Standards of Reporting Trials (CONSORT). We create an evaluation corpus from two studies on abstract reporting quality with CONSORT-abstract standards. We then evaluate the ability of different large generative language models (from the general domain or adapted to the biomedical domain) to correctly assess CONSORT criteria with different known prompting methods. Our best combination of model and prompting method achieves 85 % accuracy
@inproceedings{2025-revue-tal, title = {Évaluation de la qualité de rapport des essais cliniques avec des larges modèles de langue [Evaluation of Clinical Trials Reporting Quality using Large Language Models]}, author = {Laï-king, Mathieu and Paroubek, Patrick}, editor = {Boudin, Florian and Aizawa, Akiko}, booktitle = {Traitement Automatique des Langues, Volume 65, Numéro 2 : Numéro spécial sur le Traitement automatique de documents scientifiques [Scientific Documents Processing]}, year = {2025}, address = {France}, publisher = {ATALA (Association pour le Traitement Automatique des Langues)}, url = {https://www.atala.org/content/tal_65_2_1}, pages = {13--38}, language = {fra}, bibtex_show = true }

2024

BioNLP@ACL2024
Pre-training data selection for biomedical domain adaptation using journal impact metrics

Mathieu Laï-king, and Patrick Paroubek

In Proceedings of the 23rd Workshop on Biomedical Natural Language Processing, Aug 2024

Abs Bib PDF Code

Domain adaptation is a widely used method in natural language processing (NLP) to improve the performance of a language model within a specific domain. This method is particularly common in the biomedical domain, which sees regular publication of numerous scientific articles. PubMed, a significant corpus of text, is frequently used in the biomedical domain. The primary objective of this study is to explore whether refining a pre-training dataset using specific quality metrics for scientific papers can enhance the performance of the resulting model. To accomplish this, we employ two straightforward journal impact metrics and conduct experiments by continually pre-training BERT on various subsets of the complete PubMed training set, we then evaluate the resulting models on biomedical language understanding tasks from the BLURB benchmark. Our results show that pruning using journal impact metrics is not efficient. But we also show that pre-training using fewer abstracts (but with the same number of training steps) does not necessarily decrease the resulting model’s performance.
@inproceedings{2024-bionlp-acl, title = {Pre-training data selection for biomedical domain adaptation using journal impact metrics}, author = {Laï-king, Mathieu and Paroubek, Patrick}, editor = {Demner-Fushman, Dina and Ananiadou, Sophia and Miwa, Makoto and Roberts, Kirk and Tsujii, Junichi}, booktitle = {Proceedings of the 23rd Workshop on Biomedical Natural Language Processing}, month = aug, year = {2024}, address = {Bangkok, Thailand}, publisher = {Association for Computational Linguistics}, url = {https://aclanthology.org/2024.bionlp-1.27}, pages = {363--369}, bibtex_show = true, }

2022

n2c2 2022

KEEPHA at n2c2 2022: Track 1

Lisa Raithel, Faith W. Mutinda, Gabriel H. B. Andrade, and 13 more authors

In 2022 n2c2 Shared Task and Workshop at AMIA Fall Symposium, Nov 2022

Abs Bib PDF

Team KEEPHA submitted results for all three subtasks based on ensembles and inserting position information using different variations of transformer models. We achieve a strict F1 score of 0.95 (lenient F1 is 0.97) in medication extraction, an F1 of 0.94 (micro) and 0.87 (macro) in event classification and a combined F1 of 0.67 in context classification.

@inproceedings{2022-n2c2-keepha,
  title = {{KEEPHA at n2c2 2022: Track 1}},
  author = {Raithel, Lisa and Mutinda, Faith W. and Andrade, Gabriel H. B. and Yeh, Hui-Syuan and Nishiyama, Tomohiro and Laï-king, Mathieu and Yada, Shuntaro and Roller, Roland and Grouin, Cyril and Savary, Agata and Névéol, Aurélie and Lavergne, Thomas and Aramaki, Eiji and M{\"o}ller, Sebastian and Matsumoto, Yuji and Zweigenbaum, Pierre},
  url = {https://hal.science/hal-04297589},
  booktitle = {{2022 n2c2 Shared Task and Workshop at AMIA Fall Symposium}},
  address = {Washington, DC, United States},
  year = {2022},
  month = nov,
  hal_id = {hal-04297589},
  hal_version = {v1},
  bibtex_show = true
}