Sentiment analysis for emotional tone prediction in mental health communities with focus on comorbid conditions

In recent years, the integration of natural language processing (NLP) and machine learning has shown promise in mental health monitoring, particularly through sentiment analysis of online discourse. However, existing models often struggle to accurately detect complex emotional expressions associated with comorbid mental disorders. This study proposes a novel sentiment analysis framework designed to predict user emotional tone within online mental disorder communities, with a specific focus on comorbid conditions. The framework employs a combination of rule-based methods and deep learning techniques, including the use of TextBlob and VADER for initial sentiment extraction and DistilBERT embeddings to capture contextual nuances. These features are processed through bidirectional Recurrent Neural Networks (RNNs) to model sequential dependencies in the text. The model achieved an accuracy of 89.5%, precision of 89.7%, recall of 91.2%, F1-score of 90.4%, and an AUC-ROC score of 94.1%, demonstrating its effectiveness in capturing the intricate emotional tones present in user-generated content. This approach addresses the limitations of previous models by enhancing the detection of overlapping emotional patterns inherent in comorbid mental health conditions, thereby contributing to more accurate and nuanced sentiment analysis in mental health monitoring.