Reproduce “Explainable Depression Detection with Multi-Modalities Using a Hybrid Deep Learning Model on Social Media” on OpenResearch

About this paper

Hamad Zogan, Imran Razzak, Xianzhi Wang, Shoaib Jameel, Guandong Xu

Researchers from the University of Technology Sydney and collaborators developed MDHAN, a hybrid deep learning model that integrates multi-modal social media features with a hierarchical attention network to detect depression while providing explainable predictions. This model achieved an F1-score of 0.893 and 0.895 accuracy, effectively highlighting specific words and tweets that contributed to its decisions.

Key insights

• Integrating diverse multi-modal features with textual analysis through a hybrid deep learning architecture significantly enhances depression detection performance and provides a more holistic understanding.
• Hierarchical attention mechanisms offer crucial interpretability by highlighting specific words within tweets and particular tweets within a user's timeline that influence the model's predictions.
• Both social network features and latent topic patterns in user language are highly informative and contribute significantly to robust depression detection.

Summary by alphaXiv.