摘要
词的向量化表达是文本挖掘应用的必要前提。为了改善自编码器在词嵌入中的效果,提高文本分类的准确性,提出了一种改进的自编码器并将其用于文本分类。在传统自编码器的基础上,在隐藏层加入了一个全局调整函数,其将绝对值小的特征值调整到绝对值大的特征值上,实现了隐藏层特征向量的稀疏化。得到调整后的特征向量之后,采用全连接神经网络进行文本分类。在20news数据集上的实验结果表明,所提方法具有更好的词向量嵌入式效果,并且在文本分类中也具有更好的效果。
Vector representation of words is the premise of applications in text mining.In order to improve the effectiveness of autoencoders in words embedding and the accuracy of text lassification,this paper proposed an improved autoencoder and applied it for text classification.Based on traditional autoencoder,aglobal adjustable function is added to the latent layer,which adjusts smaller absolute values to bigger absolute values and implements the sparsity of characteristic vector in the latent layer.With the adjusted latent characteristic vector,a full connected neural network is used to classify text.The experiments on 20 news dataset show that the proposed method is more effective in words embedding,and has better performance in text classification.
作者
许卓斌
郑海山
潘竹虹
XU Zhuo-bin;ZHENG Hai-shan;PAN Zhu-hong(Information and Network Center,Xiamen Universit)
出处
《计算机科学》
CSCD
北大核心
2018年第6期208-210,240,共4页
Computer Science
基金
赛尔网络下一代互联网技术创新项目(NGII20160410)资助