一种多分类的微博垃圾用户检测方法

针对微博多类垃圾用户的检测问题,设计了一种基于模糊多类支持向量机的垃圾用户检测方法。首先,采用一对多SVM(support vector machines)的构造思想来构造多分类器,并针对每类用户的分类器重新选择训练集;然后,利用构造好的训练集来训练多分类器,经过反复调整参数,得到5个用户分类器;最后,针对多分类器的不可分样本,采用模糊聚类来进行模糊处理,即在垂直于SVM的最优分类面上定义一个改进的隶属度函数,选择最大隶属度对样本进行再分类。实验结果表明,该方法在保证垃圾用户检测效果的前提下,可以解决多分类中存在的混分和漏分问题。

A multi-classification method for detecting microblog spam users

Based on fuzzy multi-class support vector machine, a method for detecting microblog spammers is designed. Firstly, a multi-class SVM(support vector machines) is used to construct multi-classifiers, and a training set is re-selected for each type of user''s classifier. Then, the constructed training set is used to train the multi-classifier, and five user classifiers are obtained after repeated remediation. Finally, for the non-separable samples of multiple classifiers, fuzzy clustering is used to perform the fuzzy processing. An improved membership function is defined on the optimal classification plane perpendicular to the SVM, and the maximum membership degree is used to reclassify the samples. Experimental results show that this method can solve the problems of mixing and missing points in multi-classification under the premise of ensuring the detection effect of spammers.