A Multifeature-Oriented Parallel Sequence Network for Classical Music Classification in Social-Emotional Learning

Abstract

With the rapid advancement of artificial intelligence, automated social-emotional music (SEM) classification has attracted increasing attention due to its practical applications in streaming platforms, education, and music therapy. However, existing models often suffer from low classification accuracy and limited generalization capability. This study proposes the multifeature-oriented parallel sequence network (MOPSNet), a multifeature fusion convolutional recurrent neural network designed for SEM classification in learning-oriented scenarios. MOPSNet integrates complementary music representations to capture both spectral and temporal emotional cues. These features are processed through a hybrid architecture that consists of parallel CNN branches for multiscale feature extraction and stacked bidirectional long short-term memory layers for modeling long-term temporal dependencies. In addition, multihead attention, residual connections, and skip connections are incorporated to preserve fine-grained temporal information and enhance training stability. To support robust learning, a data augmentation strategy is employed to enhance model generalization under conditions of limited and noisy data, which are common in educational music datasets. Experimental results on the SEM dataset demonstrate the effectiveness of MOPSNet, achieving an average accuracy of 95.05% with a standard deviation of 1% and a 95% confidence interval of [94.33%, 95.76%]. These findings confirm that MOPSNet offers an effective artificial intelligence based learning technology for supporting social-emotional learning through music, providing interpretable and reliable feedback in educational and special learning contexts.