基于网络社团划分方法的多维数据聚类研究

为了解决传统聚类方法在多维数据集中聚类效果不佳的问题，提出了将网络社团划分的方法，并应用到多维数据聚类分析中。对于一个多维数据集，首先对分析对象进行特征提取，构建出每个对象的特征向量，通过计算皮尔森相关系数来度量不同特征向量之间的相似性，从而构建出一个相似性网络，采用Blondel算法对该网络进行社团划分达到聚类的效果。实验结果表明该方法可以在多维数据聚类中得到较好的聚类结果，准确率达到92.5%，优于K-means算法的75%。

Multidimensional data clustering based on network community detection method

Since the traditional clustering method has an impoverished clustering effect on the multidimensional data, this paper used the clustering method based on community detection of complex networks to achieve better results. Firstly, the method extracted the features of original data and formed the feature vectors of each object for a multidimensional data set. Then it measured the similarity between different feature vectors by calculating Pearson correlation coefficients and constructed a similarity network. Finally, it used the Blondel algorithm which detected the community of the network to achieve the clustering effect. Experimental results show that this method can get better clustering results in multidimensional data clustering with an accuracy rate 92.5%, which is better than K-means algorithm with the accuracy rate 75%.