网页文献的快速模糊聚类

Fast fuzzy clustering of web documents

基于对文献聚类的3种方法(c-means法、模糊c-means法和学习向量量化法)的统计和分析,借鉴了模糊聚类思想,尤其是用协方差矩阵来描述聚类的形状和大小,并将其应用于学习向量量化算法中。针对新的参考向量开发了模糊竞争学习模式,并用该算法成功地解决了文献聚类的难题。实验结果表明:学习向量量化算法能有效地解决文献的聚类问题,运行时间短;该算法与模糊聚类算法相比更健壮;该算法使在线文献聚类分析成为可能。

Based on statistics and analysis of three approaches for document clustering, the c-means, the fuzzy c-means and the learning vector quantization approach, this paper transfers some ideas from fuzzy clustering, in particular the use of a covariance matrix to describe the shape and the size of a cluster, to learning vector quantization. This paper also develops a fuzzy competitive learning scheme for these new reference vector parameters, and applies the algorithm to the difficult task of clustering documents. The experiments show that this approach can be used successfully for the clustering of documents with learning vector quantization leading to shorter execution times. The learning vector quantization appears to be more robust than fuzzy clustering. It enables a truly online clustering. 1 tab, 1 fig, 9 refs.