This research presents the development of a web-based music recommendation system that uses facial expression recognition to match songs with users' emotional states. Real-time facial detection and expression classification are conducted in the browser using two CNN models implemented via the face-api.js library. Each classified expression is mapped to a specific music genre, and relevant songs are retrieved using the SoundCloud API. The system was evaluated through two aspects, accuracy and user satisfaction. Accuracy was measured using a dichotomous questionnaire, with results showing that 91% of users agreed that the recommended songs reflected their current emotions. User satisfaction was also assessed using a similar questionnaire and reached 86%, indicating a high level of comfort and relevance in the user experience. Compared to previous studies that used Likert scales, this study offers a different yet equally effective evaluation approach. The findings suggest that integrating facial expression recognition into music recommendation systems can provide a practical and user-friendly way to support emotional regulation through music.
Wardhana et al. (Mon,) studied this question.