基于上下文感知自适应卷积网络的实验室文本分类

Laboratory text classification based on context⁃aware adaptive convolutional network

当前卷积神经网络模型运用在实验室长文本分类时,存在着特征提取不恰当、语义信息表示不完全而导致的分类不准确的问题。为了解决这一问题,文中通过构建上下文感知自适应卷积网络(CACN),采用多尺度卷积核对潜在特征进行融合,从而更好地提取单词特征的上下文信息。为验证该方法的有效性,在三个公开长文本数据集上对改进后的CACN文本分类方法进行了性能评估,在AG News、Yelp_F、Yelp_P三个长文本数据集上分类准确率分别达到了92.6%、65.5%、95.8%。

The current convolutional neural network model has the problem of inaccurate classification due to inappropriate feature extraction and incomplete representation of semantic information when it is applied to classification of long text in the laboratory.In order to deal with this problem,a context⁃aware adaptive convolutional network(CACN)is constructed in this paper,which uses multi⁃scale convolutional kernels to fuse potential features,so as to better extract the contextual information of word features.To verify the effectiveness of the method,the performance of the text classification method based on improved CACN is evaluated on three publicly available long⁃text datasets,and the classification accuracy of the method can reach 92.6%,65.5%and 95.8%on the three long⁃text datasets(AG News,Yelp_F and Yelp_P)respectively.