BFH-AMI at eRisk@ CLEF 2023

Ghofrane Merhbene, Alexandre Puttick, Mascha Kurpicz-Briki

Abstract: Mental health problems are a rising problem of today’s society. Methods of machine learning and natural language processing provide interesting new possibilities for psychology and psychiatry. In particular, eating disorders (ED) are widespread and can be life-threatening if untreated. This paper describes the approach to Task 3 of the eRisk 2023 challenge of the BFH-AMI team. The task concerned the prediction of patients’ answers to the Eating Disorder Examination Questionnaire (EDE-Q) based on their social media writing history. In our approach, we used a logistic regression model that was fed with a combination of user and question embeddings from the GPT-2 Large model.