关键词: K均值, 关联图, 初始聚类中心, 相似度矩阵

Abstract: Kmeans is the most typical clustering algorithm, which is widely used because it is concise, fast. As the traditional Kmeans is sensitive to initial clustering centers and the value of clustering parameter k is difficult to establish, this paper proposes an algorithm based on the partition of correlational graph. The algorithm can select initial clustering centers globally according to the distribution characteristics of the given data; the algorithm can determine the number of cluster automatically according to intensive degree of the given data. Effective experiments show that the algorithm has great accuracy and stability.

Key words: Kmeans, relation graph, initial clustering center, similarity matrix