基于CNN-RSC组合优化算法语音情感分析与研究

Speech Emotion Analysis and Research Based on CNN-RSC Combinatorial Optimization Algorithm

提出一种基于深度学习的文本情感分析方法,将整个卷积神经网络的模型作为一种自动学习器,对输入词语的预表达特征进行学习,引入深度学习领域的递归自编码作为输出层情感分类器,实现语义情感信息的深度提取.设置实验对比卷积神经网络和递归自编码模型的参数,找出了实验过程的最佳参数组合,实验对比了CNN、RSC、CNN-RSC三种不同的算法.实验结果表明:基于CNN-RSC的组合优化算法在对文本情感特征的自动学习上有着较好的效果,在准确度和训练时间以及分类性能上均优于其他两种算法.

In this paper, a text emotion analysis method based on depth learning is proposed. The whole convolution neural network model is used as an automatic learning device to learn the preexpression of the input words, and the recursive self encoding of the depth learning field is introduced as the output layer emotion classifier to realize the depth extraction of semantic emotion information. The experiment compares the parameters of the convolution neural network and the recursive self coding model, finds out the best combination of the experimental process, and compares the three different algorithms of CNN, RSC and CNN-RSC. The experimental results show that the combination optimization algorithm based on CNN-RSC has a good effect on the automatic learning of the text emotion feature. Accuracy and training time and classification performance are superior to the other two algorithms.