计算机工程 ›› 2019, Vol. 45 ›› Issue (8): 210-216,223.doi: 10.19678/j.issn.1000-3428.0051810

• 人工智能及识别技术 • 上一篇    下一篇

基于依存关系与神经网络的文本匹配模型

甄卓, 陈玉泉   

  1. 上海交通大学 计算机科学与工程系, 上海 200240
  • 收稿日期:2018-06-12 修回日期:2018-07-26 出版日期:2019-08-15 发布日期:2019-08-08
  • 作者简介:甄卓(1988-),男,硕士研究生,主研方向为自然语言处理、文本匹配;陈玉泉,副教授。
  • 基金项目:
    国家自然科学基金(61673266)。

Text Matching Model Based on Dependency Relation and Neural Network

ZHEN Zhuo, CHEN Yuquan   

  1. Department of Computer Science and Engineering, Shanghai Jiaotong University, Shanghai 200240, China
  • Received:2018-06-12 Revised:2018-07-26 Online:2019-08-15 Published:2019-08-08

摘要: 为增强文本匹配模型的文本语义捕捉能力并提高语义匹配准确度,提出一种基于词嵌入与依存关系的文本匹配模型。构建融合词语义和词间依存关系的语义表示,通过余弦均值卷积和K-Max池化操作获得描述两段文本各部分语义匹配程度的矩阵,并采用长短期记忆网络学习匹配程度矩阵与真实匹配程度之间的映射关系。实验结果表明,该模型的F1值为0.927 4,相比BM25及深度文本匹配模型准确度更高。

关键词: 文本匹配, 语义匹配, 依存关系, 词嵌入, 余弦均值卷积, K-Max池化, 长短期记忆网络

Abstract: In order to enhance the text semantic capture ability of text matching model and improve the semantic matching accuracy,a text matching model based on word embedding and dependency relation is proposed.It constructs the semantic representation of the fusion of word semantic and dependency relation between words,and obtains the matrix describing the semantic matching degree of each part of the two texts by cosine mean convolution and K-Max pooling operation.The Long-Short Term Memory(LSTM) network is used to learn the mapping relationship between the matching degree matrix and the true matching degree.Experimental results show that the F1 value of the model is 0.927 4,which is more accurate than the BM25 and deep text matching models.

Key words: text matching, semantic matching, dependency relation, word embedding, cosine mean convolution, K-Max pooling, Long-Short Term Memory(LSTM) network

中图分类号: