"TOWARDS A COMPASSIONATE ALGORITHM: AN INTELLIGENT TEXT AND EMOTIONS MINING FRAMEWORK FOR PREDICTING DEPRESSION AND SUICIDAL RISK ON SOCIAL MEDIA "

Abstract

The growing prevalence of depression and suicidal behavior, particularly among social media users, highlights an urgent need for intelligent systems that move beyond mere detection towards empathy-driven understanding. This research introduces “Towards a Compassionate Algorithm”—an intelligent text and emotion mining framework designed to identify early signs of depression and suicidal risk within social media communications. The proposed framework combines linguistic pattern analysis, affective computing, and deep learning models to interpret not only textual semantics but also underlying emotional cues. By integrating advanced natural language processing with sentiment and emotion classification layers, the model captures subtle shifts in tone, intensity, and behavioral expression often overlooked in traditional systems.A multi-phase methodology was employed, encompassing data preprocessing, balanced feature extraction, and hybrid deep neural architectures incorporating Bi-LSTM, BERT embeddings, and attention mechanisms. The framework emphasizes ethical and privacy-preserving data handling, aligning technological innovation with human-centered design. Experimental validation across multiple public mental health datasets demonstrates superior predictive accuracy and interpretability, particularly in identifying latent suicidal intent. Beyond performance metrics, the model is conceived as a step towards compassion-oriented artificial intelligence—technology that listens, learns, and supports rather than merely classifies. This research contributes to the evolution of socially responsible AI in mental health analytics and sets a foundation for real-time, ethically guided interventions in digital mental well-being.