基于相似性度量的高维聚类算法的研究

Research on High Dimensional Clustering Algorithm Based on Similarity Measure

针对高维数据相似度难以定义的问题,本文提出了一种新的高维数据聚类算法。该算法基于一个能够更准确表达高维数据对象之间相似性的度量函数,首先计算对象两两之间的相似度并得出一个相似度矩阵,然后根据该相似度矩阵自底向上对数据进行聚类分析。实验显示,该算法能够获得质量更高的聚类结果,并且不受孤立点影响,对数据输入顺序也不敏感。

Facing the difficult problem of how to define similarity measure for high dimensional data, a new high dimensional clustering algorithm is designed in this paper. This new clustering algorithm is based on a new similarity measure function, which can more accurately to express the similarity degree among the high dimensional data. The executing process of the algorithm is followed： firstly it uses the similarity measure function to compute the similarity degree for each high dimensional data to obtain the similarity matrix, and then conducts the cluster analysis based on the similarity matrix by the Bottom-up method. The experiment shows that this algorithm can improve the clustering analysis accurately and effectively, and will not be influent by the outliers. This algorithm is also insensitive to the input order of the data.