摘要
提出一种基于栈式自编码神经网络(SAE)的维吾尔语语句情感倾向分析方法。利用深度学习思想,将高维的维吾尔语语句空间特征向量变换到新的低维特征空间,学习并提取维吾尔语语句中隐含的语义特征。为提高特征对文本语义的表达,将富含词汇语义及上下文位置关系的句向量特征与情感特征组合进行融合,训练栈式自编码器,通过引入softmax层完成维吾尔语语句的情感分类。通过实验优选模型隐层层数、句向量维度,同传统的情感分类方法进行比较,实验结果表明,该方法更适用于维吾尔语语句情感倾向分析,微平均值为90.4%,宏平均值为90.5%。
A method based on stacked autoencoder（SAE）was presented to analyze the emotional tendency of Uyghur sentence.Deep learning was used to transform those high-dimensional vectors into new,low-dimensional and essential ones,and implicit semantic information was extracted.To improve the expression of semantic feature,sentence vector feature and small amount of emotional features were integrated to train a stacked autoencoder,and softmax regression was introduced to classify the Uyghur sentence.The number of hidden layers and paragraph vector dimensions were optimized through experimental model.The results of experiments show that the proposed approach achieves 90.4% of MicroF1 value and 90.5% of MacroF1 value,which outperforms some existed techniques for Uyghur sentence sentiment identification.
出处
《计算机工程与设计》
北大核心
2016年第8期2213-2217,共5页
Computer Engineering and Design
基金
国家自然科学基金项目(61563051
61262064
60963017)
国家自然科学基金重点项目(61331011)
