北京航空航天大学学报 ›› 2016, Vol. 42 ›› Issue (8): 1603-1611.doi: 10.13700/j.bh.1001-5965.2015.0507

• 论文 • 上一篇    下一篇

基于信息理论的网络文本组合聚类

王扬1,2, 袁昆1, 刘洪甫3, 吴俊杰1, 包秀国4   

  1. 1. 北京航空航天大学 经济管理学院, 北京 100083;
    2. 北京航空航天大学 机械工程及自动化学院, 北京 100083;
    3. 东北大学 工学院, 波士顿 02115;
    4. 国家计算机网络与信息安全管理中心, 北京 100029
  • 收稿日期:2015-07-30 出版日期:2016-08-20 发布日期:2016-09-01
  • 通讯作者: 包秀国,Tel.:010-82338497,E-mail:baoxiuguo@139.com E-mail:baoxiuguo@139.com
  • 作者简介:王扬,男,博士研究生。主要研究方向:应急管理。Tel.:010-82339105。E-mail:wyang@buaa.edu.cn;吴俊杰,男,博士,教授,博士生导师。主要研究方向:数据挖掘、社会舆情和社交网络分析。Tel.:010-82339983。E-mail:wujj@buaa.edu.cn;包秀国,男,博士研究生。主要研究方向:信息安全与大数据存储。Tel.:010-82338497。E-mail:baoxiuguo@139.com
  • 基金资助:
    国家自然科学基金(71531001,71322104,71171007,71471009);国家“863”计划(SS2014AA012303);中央高校基本科研业务费专项资金

Information-theoretic ensemble clustering on web texts

WANG Yang1,2, YUAN Kun1, LIU Hongfu3, WU Junjie1, BAO Xiuguo4   

  1. 1. School of Economics and Management, Beijing University of Aeronautics and Astronautics, Beijing 100083, China;
    2. School of Mechanical Engineering and Automation, Beijing University of Aeronautics and Astronautics, Beijing 100083, China;
    3. College of Engineering, Northeastern University, Boston 02115, USA;
    4. National Computer Network Emergency Response Technical Team/Coordination Center of China, Beijing 100029, China
  • Received:2015-07-30 Online:2016-08-20 Published:2016-09-01

摘要: 尽管近年来针对文本聚类问题进行了大量研究,其仍然是数据挖掘领域的一个富有挑战性的问题,特别在弱相关特征乃至噪声特征的处理上,仍然存在诸多挑战。针对这一问题提出了文本聚类的分解-组合算法框架——DIAS。该方法首先通过简单随机特征抽样将高维文本数据进行分解得到多样化的结构知识,其优点是能够较好地避免产生大量的噪声特征。然后采用基于信息理论的一致性聚类(ICC)将多视角基础聚类知识组合起来,得到高质量的一致性划分。最后通过在8个真实文本数据集上的实验,证明DIAS算法相较于其他被广泛使用的算法具有明显优势,特别在处理弱基础聚类上具有突出效果。由于在分布式计算上的天然优势,DIAS有望成为大规模文本聚类的主流算法。

关键词: 文本聚类, 分解-组合算法, 基于信息理论的一致性聚类, K-均值, 大数据聚类

Abstract: Although being extensively studied, text clustering remains a critical challenge in data mining community due to the curse of dimensionality. Various techniques have been proposed to overcome this difficulty, but the negative impact of weakly related or even noisy features is yet the hunting nightmare. Meanwhile, we should never lose sight of the explosive growth of unlimited user-generated content on social media, which is extremely sparse and poses further challenge on the efficiency issue. In light of this, a disassemble-assemble (DIAS) framework is proposed for text clustering. Simple random feature sampling is employed by DIAS to disassemble high-dimensional text data and gain diverse structural knowledge by avoiding the bulk of noisy features. Then the multi-view knowledge is assembled by fast information-theoretic consensus clustering (ICC) to gain a high-quality consensus partitioning. Extensive experiments on eight real-world text data sets are conducted to demonstrate the advantages of DIAS over some widely used methods. In particular, DIAS shows appealing merits in learning from a bulk of very weak basic partitionings. Its natural suitability for distributed computing makes DIAS become a promising candidate for big text clustering.

Key words: text clustering, disassemble-assemble algorithm, information-theoretic consensus clustering, K-means, big data clustering

中图分类号: 


版权所有 © 《北京航空航天大学学报》编辑部
通讯地址:北京市海淀区学院路37号 北京航空航天大学学报编辑部 邮编:100191 E-mail:jbuaa@buaa.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发