At the 28th Conference on Natural Language Processing (NLP2021) to be held between 28 June and 2 July, the Text Mining Challenge workshop (DEFT2021) will take place where scientific papers related to this workshop will be published. At this conference, 4 DRIT members will be involved in one of the following tasks. Indeed, three tasks have been proposed for DEFT this year, namely:
- Identification of the patient’s clinical profile;
- Automatic evaluation of student papers based on an existing reference;
- The automatic continuation of the correction after initial corrections.
Berger-Levrault has shown a strong interest in participating in the first task, the scientific challenge of diagnosing clinical cases. The training corpus has already been distributed by the organizers. The test corpus will be distributed from 17 May onwards, with results to be returned after a maximum of three days. The DRIT, through its members of researchers and engineers, will propose solutions to satisfy this need for classification.
The clinical cases to be treated are written in French (descriptions of rare clinical situations used for educational, scientific, or therapeutic purposes). The corpus used comes from a larger set of clinical cases with annotations. The clinical cases are anonymous. They cover different medical specialties (cardiology, urology, oncology, obstetrics, pulmonary, gastroenterology, etc.). They describe cases that occurred in different French-speaking countries (France, Belgium, Switzerland, Canada, African countries, tropical countries, etc.). In concrete terms, the objective of this task is as follows: for a given clinical case, we will be interested in identifying the clinical profile of the patient concerned by the type of disease of all the pathologies present in the case. This task builds on previous annotations of entities named pathologies, signs or symptoms, anatomical parts, etc.
Our Berger-Levrault group will already be present at TALN as a scientific paper has already been accepted for this year’s conference entitled:
“Automatic extraction of hyperonymy and hyponymy semantic relations in a business corpus”.
We look forward to DEFT2021 to add another publication!