A CNN–Transformer Framework (CS-FER) for Real-Time Facial Emotion Recognition and Customer Satisfaction Analytics in AI-Augmented Human Resource Management

The exponential growth of video surveillance data in large-scale organizations has created new opportunities to assess customer satisfaction beyond traditional surveys and interviews, which often suffer from bias, cost, and scalability limitations. This paper introduces CS-FER, a hybrid CNN–Transformer framework that translates facial emotion recognition into a quantifiable Customer Satisfaction Index (CSI). The system leverages convolutional layers for spatial feature extraction and transformer-based self-attention for temporal modeling, enabling robust real-time analysis of customer emotions under varying environmental conditions. Empirical evaluation across benchmark datasets (FER۲۰۱۳, AffectNet, RAF-DB) demonstrates that CS-FER achieves ۹۲.۶% sequence accuracy and ۰.۹۱ macro-F۱, outperforming conventional FER models. More importantly, the derived CSI correlates strongly with human-rated satisfaction scores (r = ۰.۷۸, R² = ۰.۶۱), validating its managerial relevance. Simulation of CSI-driven interventions further suggests tangible organizational benefits, including improved customer retention and enhanced employee responsiveness.By bridging the gap between technical precision in emotion recognition and actionable insights for human resource management, CS-FER highlights the potential of AI-augmented, human-centered management. Future research will extend this framework through multimodal data integration, edge optimization, and ethical-by-design practices to ensure scalability, transparency, and trust in real-world organizational contexts.

کلیدواژه ها:

Facial Emotion Recognition (FER) ، CNN–Transformer Hybrid Models ، Customer Satisfaction Index (CSI) ، AI-Augmented Human Resource Management ، Affective Computing in Organizations ، Real-Time Behavioral Analytics ، Multimodal Deep Learning ، Ethical AI in Large-Scale Systems