Text Classification For Early Detection of Eating Disorders and Depression in Spanish

@Article {
title = {Text Classification For Early Detection of Eating Disorders and Depression in Spanish},
keywds = {
BERT; Disorders Detection; Early Risk; Mental Disorders
}
abstract = {

This paper presents the participation of the Vicomtech NLP team in the MentalRiskES shared task about the early detection of mental disorders in Spanish comments from Telegram users. We participate in two tasks: Task 1a, related to eating disorders, and Task 2a related to depression. For both tasks we propose a set of approaches based on supervised text classifiers using Transformers. We prioritise our experimentation in building low resource demand systems with the minimum low carbon footprint. With those highlighted features, our systems are developed to detect disorders as early as possible, involving an initial phase that automatically projects stream labels at message level, since not all messages contained in a stream are equally representative of the stream class. We obtain the best ERDE5 result in depression detection (0.27) and second-best in eating disorders (0.17).

}
date = {2023-09-26},
}