电子学报 ›› 2020, Vol. 48 ›› Issue (11): 2131-2137.DOI: 10.3969/j.issn.0372-2112.2020.11.006

• 学术论文 • 上一篇    下一篇

熵权约束稀疏表示的短文本分类算法

脱婷1, 马慧芳1,2,3, 李志欣3, 赵卫中4   

  1. 1. 西北师范大学计算机科学与工程学院, 甘肃兰州 730070;
    2. 桂林电子科技大学广西可信软件重点实验室, 广西桂林 541004;
    3. 广西师范大学广西多源信息挖掘与安全重点实验室, 广西桂林 541004;
    4. 华中师范大学计算机学院, 湖北武汉 430079
  • 收稿日期:2018-07-30 修回日期:2020-07-01 出版日期:2020-11-25 发布日期:2020-11-25
  • 通讯作者: 马慧芳
  • 作者简介:脱婷 女,1990年9月出生,甘肃庆阳人.自2016年进入西北师范大学计算机科学与工程学院学习,现为硕士研究生,主要从事自然语言处理与分类算法方面研究.E-mail:nwnutuot@yeah.net;李志欣 男,1971年10月出生,广西桂林人.博士,博士生导师.现为广西师范大学计算机科学与信息工程学院教授,主要从事图像理解与机器学习等方面的研究.E-mail:lizx@gxnu.edu.cn;赵卫中 男,1981年10月出生,山东菏泽人.博士,硕士生导师,现为华中师范大学计算机学院副教授,主要从事机器学习与数据挖掘等方面研究工作.E-mail:zhaoweizhong@gmail.com
  • 基金资助:
    国家自然科学基金(No.61762078,No.61363058,No.61663004,No.61966004,No.61762079);广西可信软件重点实验室研究课(No.kx202003);广西多源信息挖掘与安全重点实验室开放基金(No.MIMS18-08);西北师范大学2019年度青年教师科研能力提升计划(No.NWNU-LKQN2019-2)

Effectively Classifying Short Texts by Entropy Weighted Constraints Sparse Representation

TUO Ting1, MA Hui-fang1,2,3, LI Zhi-xin3, ZHAO Wei-zhong4   

  1. 1. College of Computer Science and Engineering, Northwest Normal University, Lanzhou, Gansu 730070, China;
    2. Guangxi Key Laboratory of Trusted Software, Guilin University of Electronic Technology, Guilin, Guangxi 541004, China;
    3. Guangxi Key Lab of Multi-Source Information Mining and Security, Guangxi Normal University, Guilin, Guangxi 541004, China;
    4. School of Computer Central China Normal University, Wuhan, Hubei 430079, China
  • Received:2018-07-30 Revised:2020-07-01 Online:2020-11-25 Published:2020-11-25

摘要: 针对短文本特征稀疏性问题,提出一种熵权约束稀疏表示的短文本分类方法.考虑到初始字典维数较高,首先,利用Word2vec工具将字典中的词表示成词向量形式,然后根据加权向量平均值对原始字典进行降维.其次,利用一种快速特征子集选择算法去除字典中不相关和冗余短文本,得到过滤后的字典.再次,基于稀疏表示理论在过滤后的字典上,为目标函数设计一种熵权约束的稀疏表示方法,引入拉格朗日乘数法求得目标函数的最优值,从而得到每个类的子空间.最后,在学习到的子空间下通过计算待分类短文本与每个类中短文本的距离,并根据三种分类规则对短文本进行分类.在真实数据集上的大量实验结果表明,本文提出的方法能够有效缓解短文本特征稀疏问题且优于现有短文本分类方法.

关键词: 短文本分类, 词向量, 熵, 稀疏表示

Abstract: Aiming at the problem of short text feature sparsity,a short text sparse representation classification method based on entropy weighted constraint is proposed.Considering that the initial dictionary dimension is high,firstly,the word in the dictionary is represented as a word vector form via using the Word2vec tool,and then the original dictionary is reduced according to the average weighted vectors.Secondly,a fast feature subset selection algorithm is adopted to remove the irrelevant and redundant short texts in the dictionary,and the filtered dictionary can then be obtained.Thirdly,based on the sparse representation theory,an improved entropy-weighted sparse representation method is designed for the objective function,and the Lagrange multiplier method is introduced to obtain the optimal value of the objective function,thus the subspace of each class is obtained.Finally,the distance between the short text to be classified and the short text in each class is calculated under the subspace,and the short text is classified according to three classification rules.A large number of experimental results on real data sets show that the proposed method can effectively alleviate the short text feature sparse problem and exhibits better performance than the existing short text classification methods.

Key words: short text classification, word embedding, entropy, sparse representation

中图分类号: