研究	情感维度	准确率(%)	方法
Koelstra et al. [28]	high/low arousal	62.00	形态分量分析的特征提取方法，模态融合
	high/low valence	57.60
	liking	55.40
Salma Alhagry [21]	high/low arousal	85.65	双层LSTM网络
	high/low valence	85.45
	liking	87.99
Atkinson and Campos [30]	high/low arousal	73.06	基于最大相关最小冗余方法(mRMR)的特征选择，支持向量机(SVM)
Atkinson and Campos [30]	high/low valence	70.90	基于最大相关最小冗余方法(mRMR)的特征选择，支持向量机(SVM)
Yoon and Chung [31]	high/low arousal	73.06	基于贝叶斯定理的概率分类器，基于感知器收敛算法的监督学习
Yoon and Chung [31]	high/low valence	70.90	基于贝叶斯定理的概率分类器，基于感知器收敛算法的监督学习
Ahmet Mert [32]	high/low arousal	75.00 ± 7.48	基于多元经验模态分解的特征提取，人工神经网络
Ahmet Mert [32]	high/low valence	72.87 ± 4.68	基于多元经验模态分解的特征提取，人工神经网络
Ruo-Nan Duan, jia-Yi Zhu [18]	Positive/negative	84.22	支持向量机(SVM) K-近邻(KNN)
Chen Wei and Lan-lan Chen [24]	Positive/negative/ neutral	83.13 ± 1.67	SRU神经网络集成学习
This study	high/low arousal	87.60	多通道集成学习方法，LSTM网络
	high/low valence	90.52
	High/low dominance	68.17
	liking	80.73