BERT-Based Emotion Detection and Real-Time Content Recommendation System for Digital Mental Health Support

Abstract—Artificial intelligence (AI) in mental health has received great attention lately, especially within the realm of emotion detection and personalized support. In this paper, we introduce a BERT-based emotion detection system and demonstrate how user emotions can be used to better recommend relevant content based on their feelings at a specific moment. This contrasts with systems like Woebot and Tess, which rely on pre-scripted responses rooted in cognitive-behavioral therapy (CBT) but are inadequate for providing more extensive emotional responses that users might communicate. We utilize BERT to capture these nuances and external APIs like YouTube to offer personal video content that reflects users' emotions inside of real-time content recommendations. By capturing sentiment and guiding users through their journey, we build a dynamic recommendation engine that consistently adjusts to the changing emotions of a user, personalizing and keeping recommendations relevant on an ongoing basis. Moreover, the system complies with General Data Protection Regulation (GDPR) guidelines for privacy and data protection. Compared to conventional emotion detection models, the proposed solution improves detection precision and robustness, offering more scalable, real-time emotional well-being support. Keywords— Emotion-Based Content Recommendation, Bidirectional Encoder Representations from Transformers (BERT)), Cognitive-Behavioral Therapy (CBT), Emotional Support, Dynamic Recommendation Engine, General Data Protection Regulation (GDPR), Real-Time Adaptation, Scalability, Ethical Considerations, Digital Mental Health Support.