基于全覆盖粒计算的K-medoids文本聚类算法

K-medoids text clustering algorithm based on full covering GrC

传统K-medoids聚类算法随机选取初始聚类中心,存在迭代次数增加、聚类结果波动较大的问题,因此提出基于全覆盖粒计算的K-medoids文本聚类算法。该算法定义了全覆盖平均粒度重要性的概念。首先对文本进行Single-Pass粗聚类,利用全覆盖粒度重要性和平均粒度重要性从粗聚类结果中产生初始聚类中心候选集,再基于密度和最大最小距离法则从候选集中选出初始聚类中心。通过实验验证,该算法的聚类迭代次数明显减小,聚类质量明显提高。

The traditional K-medoids clustering algorithm selects the initial clustering center randomly,and has the problems of iteration number increase and large clustering results fluctuation,so the K-medoids text clustering algorithm based on full covering granular computing(GrC)is proposed.The concept of the full coverage average granularity importance is defined in the algorithm.The Single.Pass rough clustering is performed for the text,then the candidate sets of initial clustering center are generated from the rough clustering results by means of full covering granularity importance and average granularity importance,and the initial clustering center is selected from the candidate sets on the basis of density and max.min distance rule.The experimental results show that the clustering iteration number of the algorithm is reduced,and the clustering quality is improved obviously.