聚类初始中心点选取研究 - 南京师范大学学报》（自然科学版.pdf

33 4 () Vol. 33 No. 4 2010 12 JOURNAL OF NANJING NORMAL UNIVERSITY ( Natural Science Edition) Dec, 2010 聚类初始中心点选取研究 杨天霞, 王治和, 王 华, 王凌云 (, 730070) [ ] , K- , H uffm an ) ) ) K-SPAM ( K-means algorithm of sequence attern m ining based on theH uffm an M ethod). , / 0/ 0, . [ ] K-, , Hu ffm an, , [ ] TP391 [] A [ ] 1001-4616( 2010) 04-0161-05 Research of Clustering Initial Center Selection Yang T ianx ia, W ang Zh ihe, W ang H ua, W ang L ingyun ( College ofM athematics and Inform ation Science, Northw estNorm alUniversity, Lanzhou 730070, China) Abstract:The a er stud ied the roblem of reclustering and red iscovering in the sequence database on the basis of the resu lts of sequential attern m ining. A mi ing at this shortcom ing that it could lead to the instab ility of clustering results to select random ly the initial focal oints in the existing K-m eans clustering algorithm, an in itial center selection algorithm nam ed K-SPAM ( K-m eans algorithm of sequence attern m in ing based on theH uffm an M ethod) algorithm was ro osed. It was based on Hu ffm an idea. The algorithm cou ld reduce robability of local o tmi um to a certain extent. M oreover, a h ighly efficient/ and0 and / or0 o erators w ere ado ted to calculate smi ilarity betw een airs of sequences. To do so could greatly mi rove the execution efficiency of the algorithm. K ey words: K-m eans, sequential atterns, H uffm an tree, clustering, in itial cen ter [ 1] 1995Agraw al Srikant , , , . , . KDD, W eb. [ 2] M acQueen J B 1967K-