Detecting Anorexia Nervosa through Emotional Dynamics in Social Media

Mental health conditions affect millions of people worldwide, often disrupting their thoughts and behaviours. Detecting these issues early is both challenging and essential, as timely intervention can prevent the situation from worsening. One promising approach is to observe how individuals communicate—particularly through what they write and the emotions they express on social media. This study explores two computational methods designed to capture emotional patterns and changes in social media posts. To evaluate these methods, focused on individuals with Depression and Anorexia. The results show that both the presence and fluctuation of expressed emotions can reveal meaningful insights about users affected by these conditions. Moreover, matching the best-known approach for detecting depression and coming very close (just 1% behind) to the top-performing method for anorexia. In addition to strong performance, these emotion-based representations offer the benefit of being trust the model’s decisions.. Keywords—Mental health, Emotion analysis, social media, Depression detection, Anorexia detection, Computational modelling, Emotion variability, Affective computing, Interpretability, Machine learning.