Leveraging Emotional Intelligence Metrics and NLP-Driven Sentiment Analysis for Predictive Workplace Mental Health Monitoring

Emotion recognition from user-generated text is vital for mental health monitoring, organizational well-being, and social media analytics. We propose a multi-label classification framework leveraging a fine-tuned BERT model to detect overlapping emotional states - joy, sadness, anger, and fear in social media posts. We curated and annotated a Reddit dataset spanning eight emotions and employed transfer learning to adapt a pre-trained BERT encoder. The system architecture, defined via UML diagrams, comprises modules for tokenization, classification, and an inference workflow exposed through a Flask API. Training and validation were conducted with the Hugging Face Transformers library on GPU-accelerated Google Colab, yielding an average F1-score of 82% across all labels. To demonstrate real-world applicability, we integrated the proposed model into a React-based interface that provides real-time emotion score visualizations. Load testing under concurrent users confirmed robust scalability and responsiveness. Case studies in healthcare screening, HR analytics, and digital well-being illustrate the proposed work potential to enhance emotion-aware services. By bridging emotional intelligence metrics and advanced sentiment analysis, the proposed framework establishes a scalable, interpretable foundation for emotion recognition in organizational contexts.