摘要
针对群体聚类算法中,一般以群体成员偏好矢量的相似度作为相聚依据,但这类方法通常不能保证群体聚类后聚集的一致性的问题。提出了成员与成员集的相似度概念,给出了基于聚集一致性的成员与聚集相聚的条件,描述了一个改进的群体聚类启发式算法。同时,还定义了群体及聚集一致性的偏差指标和相对偏差指标,用以评估聚类结果。实例测试表明,该算法有较好的聚类性能和较低的一致性偏差指标。
The common method in group clustering algorithms is that the group members can be joined according to the similarity between the preference vectors of members, but this method can usually not ensure the coherence of the cluster set after group clustered. This paper puts forward the concept of the similarity between member and member set, gives the condition which can hold the coherence of cluster set when a member joined to the cluster set, and describes an improved heuristic algorithm for group clustering. Further more, this papers defines the deviation index and relative deviation index with regard to the coherence of group and cluster set, which used to evaluate the clustering issue. The result tested with instances shows that the algorithm described in this paper has better clustering performance and lower coherent deviation index.
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2007年第3期472-474,共3页
Systems Engineering and Electronics
基金
湖南省交通厅项目(200610)
长沙理工大学2005年度基金(05XXJS006)资助课题
关键词
算法
群体聚类
相似度
一致性
偏差
algorithm
group clustering
similarity
coherence
deviation
