一种基于EVS相似度的邮件社区聚类方法

聚类方法的核心是如何度量事物间的邻近性。介绍了邮件特征的向量表示形式、构建了邮件特征矩阵,并使用变形后的极值分布函数模型拟合了邮件间通信特征信息;在此基础上提出了一个新的邻近性度量方法(ex-treme value distribution similarity,EVS),用以指导邮件社区划分;使用微聚类-宏聚类邮件社区划分算法验证了该方法的有效性。实验表明,在测试数据集上,相比余弦、PCC等经典的邻近性度量方法,以EVS作为划分依据的邮件社区划分算法能够更加有效地发现高质量的邮件社区。

New email community clustering method based on EVS similarity

Proximity measurement between objects is a key problem of the clustering method. The email feature vector was introduced, and the email feature matrix was constructed. The information of email features was fitted by the model of the transformed extremal value distribution function. Based on this, EVS (extreme value distribution similarity) was proposed fur email community clustering. The effectiveness of the new measurement was verified by the micro-macro clustering algorithm. Experiments show that compared to cosine-based similarity and Pearson correlation coefficient, the algorithm using the new proposed similarity measurement can identify higher quality communities.