基于k_means的中文文本聚类算法.pdf

2010 International Conference on Services Science, Management and Engineering Chinese text clustering algorithm based k-means Mingyu Yao ，Dechang Pi Xiangxiang Cong College of Information Science and Technology, Nanjing School of Information Science Engineering, East China University of Aeronautics and Astronautics University of Science and Technology Nanjing, Jiangsu, 210016, PR China Shanghai, 200237, PR China {my_yao@163.com, dc.pi@} Abstract—Text clustering is an important means and method in text mining. The process of Chinese text clustering based on k-means was emphasized, we found that new center of a cluster was easily effected by isolated text after some experiments. Average similarity of one cluster was used as a parameter, and multiplied it with a modulus between 0.75 and 1.25 to get the similarity threshold value, the texts whose similarity with original cluster center was greater than or equal to the threshold value ware collected as a candidate collection, then updated the cluster center with center of candidate collection. The experiments show that improved method averagely increased purity and F value about 10 percent over the original method. Keywords-text cluster; k-means; Chinese text 基于k-means的中文文本聚类算法 1 1 2 姚明宇 ，皮德常 ，丛湘香 1.南京航空航天大学 信息科学与技术学院，江苏 南京，中国，210016 2.华东理工大学 信息科学与工程学院，上海，中国，200237 1. my_yao@163.com, 2. dc.pi@ 【摘要】文本聚类是文本挖掘的重要手段和方法。重点分析了基于 k-means 的中文文本聚类主要过程并进行实 验，发现新的聚类中心容易受孤立文本的影响。用簇的平均相似度作为参数，乘以范围在 0.75 到 1.25 之间的 系数生成相似度阈值，用与原簇中心相似度大于等于该阈值的部分文本