AI-powered mental health application with data privacy preservation.

Area of Science:

• Digital Mental Health
• Artificial Intelligence
• Natural Language Processing

Background:

• Increasing prevalence of anxiety disorders, emotional conditions, and stress in modern society.
• High demand for digital mental health technologies offering support.
• Need for privacy-preserving emotion recognition systems.

Purpose of the Study:

• Propose an AI-driven framework for emotion recognition in user text.
• Integrate Natural Language Processing (NLP) with privacy-protective systems.
• Enhance digital mental health support through accurate emotion detection.

Main Methods:

• Utilizing BERT-base-uncased and ModernBERT Large transformer models for multi-label emotion classification.
• Implementing AES-256 encryption, Supabase authentication, and role-based access control for data security.
• Analyzing class imbalance in emotions and evaluating precision-recall trade-offs.

Main Results:

• ModernBERT demonstrated rapid convergence and improved sensitivity to sarcasm and mixed emotional states.
• Developed a privacy-first architecture for integration with digital therapy and clinical decision software.
• Designed an emotion feedback loop for contextual responses and personalized mindfulness.

Conclusions:

• The proposed AI framework effectively recognizes emotions in text using advanced NLP models.
• The privacy-first architecture ensures data security and facilitates integration with existing digital health solutions.
• The system offers potential for enhanced, personalized digital mental health interventions.