基于自适应簇中心选择的文本聚类算法研究  

An Adaptive Cluster Center Selection Algorithm

作　　者：翟东海[1,2] 聂洪玉[1] 崔静静[1] 杜佳[1] 

机构地区：[1]西南交通大学信息科学与技术学院,四川成都610031 [2]西藏大学工学院,西藏拉萨850000

出　　处：《成都信息工程学院学报》2013年第6期617-622,共6页Journal of Chengdu University of Information Technology

基　　金：国家语委"十二五"科研规划资助项目(YB125-49);教育部科学技术研究重点资助项目(212167);中央高校基本科研业务费专项资金科技创新资助项目(SWJTU12CX096);国家级大学生创新创业训练计划资助项目(201210694017)

摘　　要：为解决传统的K-means算法需要人工确定K值和随机选取初始簇中心容易陷入局部最优的问题,提出自适应簇中心选择算法。首先将任意选取的一篇文档和与其距离最远的文档作为初始簇中心聚类得到2个大类并重新计算簇中心,然后,找出与新的簇中心距离大于设定阈值的文档并依据文档距离判断是否需要增加新的类别,迭代上述过程确定聚类簇中心及类别数。实例验证结果表明,提出的算法与改进的K-means算法相比,在聚类结果的质量和算法收敛的速度上都有明显的改善。To solve problems of manual K value determination and initial cluster center random selection in original K- means is prone to local optimal, an adaptive cluster center selection algorithm is proposed in this paper. Firstly, select a document and the another one is of the farthest from it as the two initial centers to cluster. The two clusters are used to recalculate their new cluster centers. Secondly, those documents whose distances from the two new cluster centers are above the threshold are selected to determine whether new cluster center is needed. Finally, the above- mentioned procedure iterates to determine the all cluster centers and their number K. The experimental results show that compared with the improved K-means algorithm the proposed method can achieve high clustering quality and satisfactory convergence speed.

关 键 词：海量数据挖掘 初始簇中心 文档距离 K-MEANS算法 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]