基于改进卷积神经网络的多标签文本自动化分类研究

Automatic classification of multi-label text based on improved convolutional neural network

为实现多标签文本自动分类,有效管理与检索文本信息,为后续的数据挖掘作好准备,提出基于改进卷积神经网络的多标签文本自动化分类方法。在卷积神经网络卷积层前、后分别加入注意力机制,使网络的注意力集中在多标签文本重要特征上,达到改进卷积神经网络的目的,利用改进后的网络提取多标签文本的文本特征;通过附加类别标签的LDA模型获得词-标签概率数据,将其输入至双向长短期记忆网络模型中,得到多标签文本各个标签间的相关性,并结合改进卷积神经网络提取多标签文本词-标签特征;将两种特征实施拼接操作后,输入至训练好的全连接神经网络中,完成多标签文本的分类与输出。实验证明:该方法可以有效实现多标签文本自动化分类,应用的网络模型较为合理,面对不同语言的多标签文本也能够较好地完成自动化分类。

A multi-label text automatic classification method based on improved convolutional neural network was proposed to realize automatic classification of multi-label text,effectively manage and retrieve text information,and prepare for subsequent data mining.The attention mechanism is added before and after the convolutional neural network convolutional layer to make the network focus on the important features of the multi-label text,so as to improve the convolutional neural network and extract the text features of the multi-label text by using the improved network.Word-label probability data was obtained by the LDA model with additional category tags,which was input into the bidirectional long-and short-term memory network model to obtain the correlation between each tag in the multi-label text,and combined with the improved convolutional neural network to extract the multi-label text word-label features.After concatenating the two features,they are input into the trained fully connected neural network to complete the classification and output of the multi-label text.The experimental results show that the proposed method can effectively realize the automatic classification of multi-label text,and the network model applied is reasonable.It can also complete the automatic classification of multi-label text in different languages.