计算机工程 ›› 2018, Vol. 44 ›› Issue (11): 190-196.doi: 10.19678/j.issn.1000-3428.0048715

• 人工智能及识别技术 • 上一篇    下一篇

基于VDCNN与LSTM混合模型的中文文本分类研究

彭玉青,宋初柏,闫倩,赵晓松,魏铭   

  1. 河北工业大学 人工智能与数据科学学院,天津 300401
  • 收稿日期:2017-09-18 出版日期:2018-11-15 发布日期:2018-11-15
  • 作者简介:彭玉青(1969—),女,教授,主研方向为智能信息处理、计算机视觉;宋初柏、闫倩、赵晓松、魏铭,硕士研究生
  • 基金项目:

    河北省自然科学基金重点项目(F2016202144);河北省自然科学基金面上项目(F2017202145)

Research on Chinese Text Classification Based on Hybrid Model of VDCNN and LSTM

PENG Yuqing,SONG Chubai,YAN Qian,ZHAO Xiaosong,WEI Ming   

  1. School of Artificial Intelligence,Hebei University of Technology,Tianjin 300401,China
  • Received:2017-09-18 Online:2018-11-15 Published:2018-11-15

摘要:

自然语言在结构上存在一定的前后依赖性,且将中文文本直接转化为向量时会使维度过高,从而导致现有文本分类方法精度较低。为此,建立一种超深卷积神经网络(VDCNN)与长短期记忆网络(LSTM)相结合的混合模型。通过VDCNN的深度结构来提取文本向量的特征,利用LSTM具有存储历史信息的特点提取长文本的上下文依赖关系,同时引入词嵌入将文本转换为低维度向量。在Sogou语料库和复旦大学中文语料库上进行实验,结果表明,相对CNN+rand、LSTM等模型,该混合模型可以有效提升文本分类的精确率。

关键词: 文本分类, 卷积神经网络, 长短期记忆网络, 词嵌入, 深度残差网络

Abstract:

Due to the structural dependence of natural language,and the high dimensionality when Chinese text is directly transformed into vector,the accuracy of existing text classification methods is low.To solve this problem,a hybrid model of Very Deep Convolution Neural Network(VDCNN) and Long Short-Term Memory network(LSTM) is proposed.The depth structure of VDCNN is used to extract the features of text vectors,the context dependence of long text is extracted by using LSTM’s feature of storing historical information,and word embedding is introduced to transform text into low-dimensional vector.Experimental on Sogou corpus and Fudan University Chinese corpus show that,the hybrid model can effectively improve the accuracy of text classification compared with CNN+rand and LSTM.

Key words: text classification, Convolution Neural Network(CNN), Long Short-Term Memory network(LSTM), word embedding, deep Residual Networks(ResNets)

中图分类号: