Valence-arousal classification of emotion evoked by Chinese ancient-style music using 1D-CNN-BiLSTM model on EEG signals for college students
Ruoyu Du,Shujin Zhu,Huangjing Ni,Tianyi Mao,Jiajia Li,Ran Wei
DOI: https://doi.org/10.1007/s11042-022-14011-7
IF: 2.577
2022-10-04
Multimedia Tools and Applications
Abstract:During the COVID-19 pandemic, young people are using multimedia content more frequently to communicate with each other on Internet platforms. Among them, music, as psychological support for a lonely life in this special period, is a powerful tool for emotional self-regulation and getting rid of loneliness. More and more attention has been paid to the music recommender system based on emotion. In recent years, Chinese music has tended to be considered an independent genre. Chinese ancient-style music is one of the new folk music styles in Chinese music and is becoming more and more popular among young people. The complexity of Chinese-style music brings significant challenges to the quantitative calculation of music. To effectively solve the problem of emotion classification in music information search, emotion is often characterized by valence and arousal. This paper focuses on the valence and arousal classification of Chinese ancient-style music-evoked emotion. It proposes a hybrid one-dimensional convolutional neural network and bidirectional and unidirectional long short-term memory model (1D-CNN-BiLSTM). And a self-acquisition EEG dataset for Chinese college students was designed to classify music-induced emotion by valence-arousal based on EEG. In addition to that, the proposed 1D-CNN-BILSTM model verified the performance of public datasets DEAP and DREAMER, as well as the self-acquisition dataset DESC. The experimental results show that, compared with traditional LSTM and 1D-CNN-LSTM models, the proposed method has the highest accuracy in the valence classification task of music-induced emotion, reaching 94.85%, 98.41%, and 99.27%, respectively. The accuracy of the arousal classification task also gained 93.40%, 98.23%, and 99.20%, respectively. In addition, compared with the positive valence classification results of emotion, this method has obvious advantages in negative valence classification. This study provides a computational classification model for a music recommender system with emotion. It also provides some theoretical support for the brain-computer interactive (BCI) application products of Chinese ancient-style music which is popular among young people.
computer science, information systems, theory & methods,engineering, electrical & electronic, software engineering