Computer Engineering and Applications ›› 2017, Vol. 53 ›› Issue (4): 90-97.DOI: 10.3778/j.issn.1002-8331.1507-0176

Previous Articles     Next Articles

Detection of hype groups based on mining maximum frequent itemsets in Microblogs

LIU Yan, ZHANG Jin, CHEN Jing, YIN Meijuan, ZHANG Weili   

  1. State Key Laboratory of Mathematical Engineering and Advanced Computing, Zhengzhou 450002, China
  • Online:2017-02-15 Published:2017-05-11

基于最大频繁项集挖掘的微博炒作群体发现方法

刘  琰,张  进,陈  静,尹美娟,张伟丽   

  1. 数学工程与先进计算国家重点实验室,郑州 450002

Abstract: In recent years, the hype accounts in Microblogs rise as a new force, using illegal means to carry out the network public relations activities, which has seriously disturbed the normal order of the Internet. The traditional detection of hype accounts mainly uses methods based on feature analysis, ignoring that hype accounts are strongly organizational and planning, which is difficult to find the concealed ones. In view of the above problems, fully considering the group characteristics that hype accounts often participate in hype microblogs together, the problem of hype groups detection is transformed into the problem of mining maximum frequent itemsets, and a method based on mining maximum frequent itemsets for the detection of hype groups is proposed, which can find accounts groups who have participated in hype microblogs together in many times. According to the research background and the characteristics of transaction database, a new algorithm based on iterative intersection is proposed to improve the efficiency of mining maximum frequent itemsets, which uses a selection strategy based on binary search algorithm to reduce the transaction database, and uses a variety of ways to reduce the times of intersection between transactions. Finally, the performance of IIA algorithm is evaluated by experiments, and experiments are conducted on a real dataset from Sina Weibo, the experiments results show that this method can find highly concealed hype accounts that can’t be identified by traditional methods based on feature analysis, with the accuracy rate of up to 90%.

Key words: data mining, microblog, hype groups, maximum frequent itemsets

摘要: 近年来微博炒作账户异军突起,采用违规手段开展网络公关活动,严重扰乱了正常的互联网秩序。传统的炒作账户发现主要采用特征分析方法,忽视了炒作账户的组织性和策划性,难以发现隐蔽性高的炒作账户。针对以上问题,充分考虑到炒作账户共同参与微博炒作的群体特性,将炒作群体发现问题转化为挖掘最大频繁项集问题,提出了一种基于最大频繁项集挖掘的炒作群体发现方法,能够找出多次共同参与炒作微博传播的账户群体。为了提高最大频繁项集挖掘的效率,结合研究背景以及事务数据库的特点,提出了一种基于迭代交集的最大频繁项集发现算法,采用基于二分查找的最大频繁候选项集筛选策略对事务数据库进行缩减,并利用多种方式减少事务间取交集的次数。最后通过实验对IIA算法的性能进行了评估,并在真实的新浪微博数据集上验证了炒作群体发现方法的有效性,实验结果表明利用该方法发现的炒作群体准确率高于90%,而且能发现传统特征分析方法难以识别的隐蔽炒作账户。

关键词: 数据挖掘, 微博, 炒作群体, 最大频繁项集