Data Augmentation for Multi-Class Eating Disorders Text Classification

In this study, we tackle the challenge of detecting Eating Disorders (EDs) in German text, a relatively unexplored area in natural language processing (NLP) for mental health. In this project, we developed a manually annotated German dataset from YouTube comments. To address the class distribution imbalance, we employed back translation as a data augmentation technique. This process significantly enhanced the dataset’s utility. Through a comprehensive grid search, we identified a Support Vector Machine (SVM) model as the most effective, achieving an average F1-score of 0.83. Our findings not only contribute to the research field of ED detection in German but also demonstrate the effectiveness of innovative data augmentation techniques in managing class imbalances in natural language processing.