周枫,李荣雨.基于BGRU池的卷积神经网络文本分类模型[J].计算机科学,2018,45(6):235-240
基于BGRU池的卷积神经网络文本分类模型
Convolutional Neural Network Model for Text Classification Based on BGRU Pooling
投稿时间:2017-05-03  修订日期:2017-09-08
DOI:10.11896/j.issn.1002-137X.2018.06.042
中文关键词:  深度学习,卷积神经网络,双向门控循环单元,文本分类
英文关键词:Deep learning,Convolutional neural network,Bi-directional gated recurrent unit,Text classification
基金项目:本文受江苏省高校自然科学基金资助
作者单位E-mail
周枫 南京工业大学计算机科学与技术学院 南京211816  
李荣雨 南京工业大学计算机科学与技术学院 南京211816 alwayslry@sina.com 
摘要点击次数: 162
全文下载次数: 126
中文摘要:
      针对深度学习在处理文本分类问题时存在的适应度小、精确度较低等问题,提出一种采用双向门控循环单元(BGRU)进行池化的改进卷积神经网络模型。在池化阶段,将BGRU产生的中间句子表示与由卷积层得到的局部表示进行对比,将相似度高的判定为重要信息,并通过增大其权重来保留此信息。该模型可以进行端到端的训练,对多种类型的文本进行训练,适应性较强。实验结果表明,相较于其他同类模型,提出的改进模型在学习能力上有较大优势,分类精度也有显著提高。
英文摘要:
      Aiming at the problem that deep learning has the disadvantages of small adaptability and low precision when it solves the problem of text classification,this paper proposed a convolution neural network model based on bi-directional gated recurrent unit (BGRU) and convolution layer pooling.In the pooling stage,the intermediate sentence gene-rated by BGRU is represented as a local representation obtained from the convolution layer,the representation of high similarity is judged to be important information,and the information is retained by increasing its weight.The model can give end-to-end training and train multiple types of text,and it has good adaptability.The experimental results show that the proposed model has greate advantage compared with other similar models,and the classification accuracy is also improved significantly.
查看全文  查看/发表评论  下载PDF阅读器