基于半监督学习的K-均值聚类算法研究

Semi-supervised learning based on K-means clustering algorithm

定义了一个欧氏距离和监督信息相混合的新的最近邻计算函数,从而将K-均值算法很好地应用于半监督聚类问题。针对K-均值算法初始质心敏感的缺陷,用粒子群算法的搜索空间模拟聚类的欧氏空间,迭代搜索找到较优的聚类质心,同时提出动态管理种群的策略以提高粒子群算法搜索效率。算法在UCI的多个数据集上测试都得到了较好的聚类准确率。

This paper constructed a new classified function which mixed Euclidean distance with supervising information. Taking into account that K-means algorithm was sensitive to the initial center, used search space of particle swarm algorithm was used to simulate the clustering Euclidean space to find a better cluster center of clustering. At the same time, brought up a strategy of species dynamic management to improve the efficiency of particle swarm optimization search. The algorithm got a good clustering accuracy on a number of UCI testing data sets.