中文摘要：

本文提出了一种用于聚类分析的加权聚类算法,通过利用拉普拉斯权,将聚类对象之间的结构信息自动转换为对象的权重.由于拉普拉斯权能够描述数据的邻域结构,从而能够更好的聚类.该加权聚类算法在性能上比经典聚类算法有较大改进,还具有对孤立点鲁棒、适合类别不平衡数据聚类、对聚类个数不敏感等优点.人工数据集以及UCI标准数据集上的实验证实了本文算法的可行性和有效性.

英文摘要：

In this paper,we propose a novel weighted clustering algorithm based on Laplacian weight,which can automatically transform the structure information between clustering objects into weights of objects.Because Laplacian weight can indicate the neighborhood structure of original data set,better clustering is achieved.Performed on conventional C-means or fuzzy C-means methods,the proposed Laplacian weighting scheme can effectively improve the clustering performance.In addition,the new algorithm achieves some extra advantages such as robustness to outliers, suitability for class-imbalance data clustering and insensitivity to number of clusters, etc. Experimental results on artificial datasets and UCI machine learning repository validate the effectiveness of the proposed algorithm.