The study aimed to identify machine learning models that could efficiently categorize tweets concerning eating disorders. Over a million tweets were collected over three months and classified using both traditional and deep learning models. The models classified the tweets based on whether they were written by people suffering from eating disorders, promoted such disorders, were informative or scientific. The bidirectional encoder representations from transformer (BERT) based models had the best performance, despite having a higher computational cost.
Publication date: 9 Feb 2024
Project Page: https://medinform.jmir.org/2022/2/e34492
Paper: https://arxiv.org/pdf/2402.05571
