Depression Detection on Social Media Posts Using BERT and Support Vector Machine

Depression is a growing mental health challenge, affecting countless individuals worldwide. As people increasingly express their mental states on social media platforms like Twitter and Reddit, there is an opportunity to leverage this data to identify signs of depression. This study investigates the potential of social media text for detecting depression using machine learning techniques. We used Kaggle-sourced Twitter and Reddit datasets, each containing labeled samples of depressive and non-depressive content. After extensive preprocessing such as text cleaning, tokenization, and BERT embedding generation we trained Support Vector Machine (SVM) models with various kernel functions. Our results show Twitter data significantly enhances model performance, achieving 94.4% accuracy with the polynomial kernel while 87.5% on twitter dataset with Radial basis function (RBF). These findings underscore the potential for accurate depression detection, suggesting its value for early intervention and use of social media data for depression detection.