摘要
聚类集成可以有效提高传统聚类算法的精度,其关键问题在于如何根据聚类成员提供的信息获得更加优越的聚类结果。设计一种聚类集成算法,它结合K均值算法与基于拉普拉斯矩阵的谱聚类算法,充分利用聚类成员提供的属性信息与关系信息。为了降低算法计算复杂度,通过代数变换方法有效避免了大规模矩阵的特征值分解问题。在多组真实数据集上的实验结果表明,提出的算法优于其他聚类集成算法。
Cluster ensemble can effectively improve the accuracy of traditional clustering algorithms, whereupon the key problem lies in how to yield final superior result according to the information provided by cluster members. In this paper, we design a cluster ensemble method which makes full use of attribution information and relation information provided by cluster members and integrates the algorithms of K-means and Laplacian matrix-based spectral clustering. To reduce the computational complexity, we propose an algebraic transformation to avoid the eigenvalue decomposition problem of large-scale matrix. Results of experiment on several groups of real datasets demonstrate that the proposed algorithm outperforms other cluster ensemble algorithm.
出处
《计算机应用与软件》
CSCD
北大核心
2012年第10期69-70,140,共3页
Computer Applications and Software
基金
国家自然科学基金项目(60975042
41006057
61105057
61102105)
盐城工学院人才引进专项基金项目(XKR2011019)
