Mood-Driven Personalized Recommendation System using ERS

Mood-based recommendation systems have gained importance in improving user experience and emotional well-being. However, most existing systems provide recommendations based on instantaneous, often unstable emotion detection and fail to evaluate whether those recommendations truly help users recover. This limitation reduces personalization and long-term effectiveness. This project proposes a Mood-Driven Personalized Recommendation System based on an Emotional Recovery Signature (ERS), supported by a state-of-the-art, high-accuracy emotion AI architecture. The system detects user emotions through a robust multimodal pipeline, utilizing RetinaFace for precise face detection and alignment, alongside an EfficientNet-B4 backbone enhanced with a Convolutional Block Attention Module (CBAM) for deep facial feature extraction. To guarantee stable real-time inference, the model incorporates confidence calibration and temporal emotion smoothing, complemented by text sentiment analysis. Based on the accurately detected emotion, adaptive interventions—such as music recommendations, book suggestions, guided breathing exercises, and wellness resources—are provided. The Emotional Recovery Measurement Algorithm (ERMA) tracks emotional transitions over time, while the Adaptive Emotional Intervention Selection Algorithm (AEISA) dynamically selects the most effective intervention based on the user’s recovery pattern. Developed using Python, PyTorch, OpenCV, and Streamlit, the system delivers highly accurate real-time emotion detection, reliable recovery tracking, and tailored content. Unlike traditional mood-based applications, the proposed system leverages an advanced deep learning pipeline to optimize emotional recovery and promote mental wellness through adaptive, evidence-based personalization.

Keywords: Emotion AI, Affective Computing, Deep Learning, Personalized Recommendation System, Emotional Recovery Signature (ERS), Mental Wellness.