Participation de l’équipe TTGV à DEFT 2023 : Réponse automatique à des QCM issus d’examens en pharmacie

Blivet et al. 2023
Conference paper - CORIA-TALN’2023

Abstract

This article presents the TTGV team’s approach to the two tasks proposed for DEFT 2023: identifying the number of supposedly correct answers to a MCQ and predicting the set of correct answers among the five proposed for a given question. This article presents the different methodologies implemented, exploring a wide range of approaches and techniques to address first the distinction between questions calling for a single or multiple answers, before looking at the identification of correct answers. We will detail the different methods used, highlighting their respective advantages and limitations. We will then present the results obtained for each approach. Finally, we will discuss the limitations intrinsic to the tasks themselves and to the approaches considered in this contribution.

Keywords: classification, multiclass, multi-label, generative approaches