An AI-Based Hybrid Model to Identify People Who May Be Depressed Based on Their Anonymous Posts on Social Media
An AI-Based Hybrid Model to Identify People Who May Be Depressed Based on Their Anonymous Posts on Social Media
Shrijay Ramdas Kale¹, Dr. Santosh Gaikwad²
¹PG Student, Department of Computer Science and Applications
²Associate Professor, Department of Computer Science and Applications Faculty of Science and Technology, JSPM University, Pune, Maharashtra, India
I. ABSTRACT
Many people suffer from depression, which is a serious mental health issue. A growing number of individuals are utilizing social media to express their experiences and emotions; thus, monitoring their posts provides an opportunity to identify signs of depression in its early stages. This research proposes a Hybrid Explainable Multimodal Deep Learning Framework (HEMDL) for early detection of depressive tendencies in anonymous social media text. The framework combines text analysis using RoBERTa embeddings with time-based models (Bi-directional Long Short-Term Memory (BiLSTM) networks) and behavioral factors such as frequency of posts, mood changes over time, emoji use, and user interactions. It incorporates Explainable AI techniques (SHAP and attention-based visualizations) for transparency and trust, and supports multiple languages including English, Hindi, and Marathi. The proposed system is expected to achieve 98–99% accuracy, outperforming existing CNN, LSTM, BiLSTM, and Transformer models.
Keywords: Depression Detection, Deep Learning, Natural Language Processing, RoBERTa, BiLSTM, Explainable AI, Social Media Analysis, Mental Health Monitoring, Multimodal Learning, Anonymous Social Media.