基于句子级学习改进CNN的短文本分类方法

摘要

To improve the performance of network short text classification, a fusion method of convolution neural network (CNN) and sentence-level supervised learning was proposed.A classic CNN model was built for short text classification.The subject sentence was integrated into the CNN, the sentence-level CNN supervised learning for the input text was executed, and sentence model was built and the subject sentence was identified.The subject sentence model was given a higher weight, and the text model was constructed by weighting.Text classification was achieved through text-level CNN supervised learning.Experimental results on the two review datasets show that the proposed method has high classification accuracy.%为提高对网络短文本分类的性能, 提出一种融合卷积神经网络 (CNN) 和句子级监督学习的分类方法.构建一种用于短文本分类的经典CNN模型;将主题句融入到CNN中, 即对输入文本进行句子级CNN监督学习, 构建句子模型并识别主题句;将主题句子模型赋予较高权重, 通过加权和构建文本模型.通过文本级CNN监督学习, 实现文本分类.在两个评论数据集上的实验结果表明, 提出方法具有较高的分类准确性.