An efficient approach to detecting mental illness on online social media platforms by using multidimensional textual features.

Résumé

BACKGROUND: Mental illness detection on online social media platforms has become an increasingly critical research area. Existing approaches typically rely on a single type of textual feature, such as word embeddings or sentiment analysis, which limits their ability to capture the nuanced and multidimensional nature of user-generated content.OBJECTIVE: This study aims to improve the detection of mental illness in social media posts by introducing a model that integrates multiple dimensions of textual information, thereby enhancing classification performance across varied datasets.METHODS: We propose a novel model that combines several textual features, including word embeddings, sentiment scores, emotional tones, and topical context. The performance of our model was evaluated against multiple state-of-the-art text classification baselines using two real-world Reddit datasets that vary in size and class distribution.RESULTS: Our model demonstrated superior performance across several evaluation metrics, including precision, recall, Matthews Correlation Coefficient (MCC), and F1 score. It achieved an accuracy of 75.13% on a larger dataset and 69.49% on a smaller or more diverse dataset. These results highlight the robustness and adaptability of our model across different data conditions.CONCLUSION: The integration of multidimensional textual features significantly enhances the accuracy and generalizability of mental illness detection models. Our approach provides a promising direction for future research and practical applications in mental health monitoring on social media platforms.