Music emotion classification based on deep neural network and its application in context aware recommendation system

Abstract Music plays a vital role in shaping human emotional experiences, making music emotion classification a key component for enhancing personalized user interaction. The classification of music emotion has emerged as a powerful tool to improve personalized music recommendation systems by aligning musical content with users’ emotional states. This research presents a novel deep learning-based model for classifying music emotions and integrating these classifications into a context-aware recommendation framework. In the proposed system, music data and emotion tags are processed through audio feature extraction and metadata normalization. User context, including time, location, and event-based emotions, is collected via mobile applications. Emotion labels are encoded into the valence-arousal space, and the data is cleaned, normalized, and split for training and evaluation. Audio features are extracted using Mel-Frequency Cepstral Coefficients (MFCCs). The scalable parrot-optimized context-aware attention-based deep neural network (SPO-CAAtt-DNN) model is designed to classify music emotions. This model is trained to predict continuous emotion values for both music and user preferences. Implemented in Python, the findings demonstrate that the SPO-CAAtt-DNN approach outperforms the DL-MRF baseline on the same experimental setup, yielding superior results with accuracy, F1-score, recall, and precision consistently achieving an accuracy of 93.85% under the final experimental setting. These findings confirm that incorporating deep learning-based emotion classification with contextual information significantly enhances the quality of music recommendations.