Classification and identification of emotion of non-foreign music based on TR-Bi-LSTM emotion analysis

AbstractIntangible cultural heritage (IntCH) folk music has garnered considerable attention due to its immense potential cultural, artistic, and commercial value. With the advancement of the national economy, artistic forms have undergone unprecedented diversification, leading to a corresponding cultural impact on IntCH folk music. Unfortunately, this has also led to the loss of some of these invaluable musical traditions. In this study, we propose an improved bi-directional LSTM for accurately classifying and recognizing the sentiment of IntCH music. Our hybrid network model, TR-Bi-LSTM, incorporates an attention mechanism in the bidirectional LSTM, allowing us to enhance the sentiment analysis at crucial junctures in the music and further augment the accuracy of the model. Additionally, we employ a Dense-ResNet network to perform preliminary feature extraction of the emotion of non-foreign music, thereby strengthening the overall effectiveness of the model. Experimental results indicate that our proposed approach surpasses existing models in terms of accuracy, F1 score, and AUC evaluation index, thus making a significant contribution to the transmission and development of non-foreign music.