A Deep Learning Model for Identifying and Analyzing Eating Disorder-Related Communication on Social Media

Abstract

This research introduces a novel deep learning model to identify individuals at risk of eating disorders (ED) through social media data, particularly on Twitter. Using bidirectional encoder representations from transformers (BERT) and long short-term memory (LSTM), the model achieves exceptional performance with an F1 score of 98.19% and an accuracy of 98.37% on a dataset of 1.4 million Twitter biographies. Categorizing users into groups such as ED-users, healthcare professionals, and communicators, the study unveils distinct communication patterns. Notably, ED-users tend toward secrecy, while healthcare professionals and communicators engage more openly. This research pioneers the identification of user categories in ED-focused communication, offering valuable insights and potentially enhancing early detection and communication between individuals at risk of ED and healthcare professionals.