Emotion-Based Music Recommendation System Using Multi-Task Cascaded Convolutional Networks

Music has therapeutic potential in influencing and regulating human emotional states. Traditional music recommendation systems primarily rely on user mood preferences and song characteristics, which is a tedious task and often fails in case of real-time emotional shifts, limiting their effectiveness in mood-based music therapy applications. The proposed work explores the integration of music recommendation system with facial emotion recognition by using a webcam for capturing realtime mood from facial expressions. The captured facial expression is then analyzed using Multi-task Cascaded Convolutional Networks (MTCNN) to detect user's emotional state, which is then mapped to musical attributes such as valence, tempo, and energy using a recommendation algorithm to recommend songs that align with user's current mood. Various evaluation metrics such as accuracy, ranking quality, novelty, and diversity have been used to draw a comparative analysis among various recommendation algorithms. The results demonstrate 86% accuracy in facial-based music recommendations.