中文摘要：

提出了一种GML文档结构聚类新算法MCF_CLU.与其它相关算法不同,该算法基于闭合频繁Induced子树进行聚类,聚类过程中不需树之间的两两相似度比较,而是挖掘GML文档数据库的闭合频繁Induced子树,为每个文档求一个闭合频繁Induced子树作为该文档的代表树,将具有相同代表树的文档聚为一类.聚类过程中自动生成簇的个数,为每个簇形成聚类描述,而且能够发现孤立点.实验结果表明算法MCF_CLU是有效的,且性能优于其它同类算法.

英文摘要：

This paper presents an algorithm MCF _ CLU for clustering GML documents by structure. Different from other algorithms, it goes on clustering based on the closed frequent induced subtrees, and doesnt need comparing the similarity between trees. The closed frequent induced subtrees of all the GML documents are computed. The representative closed frequent induced subtree of every document is obtained. The documents which have the same representative tree are regarded as a cluster. During the clustering process, not only the number of clusters can be obtained automatically, but the description of the clusters can be achieved. By the way, the isolated points of the documents can be found. The experimental results show that MCF _ CLU is effective, and that its performance is superior to those of other GML clustering algorithms.