【作者】 田伟；

【导师】 王秀坤；

【作者基本信息】 大连理工大学 ， 计算机应用技术， 2009， 硕士

【摘要】 随着网络技术的飞速发展,信息大量膨胀和聚集,XML(eXtensible MarkupLanguage)作为一种常用的数据交换和传输标准,蕴含了丰富的信息。因此,对XML文档的挖掘已经成为Web挖掘一个新的研究热点。本文重点对XML文档分类方法进行了研究。XML文档的结构特征是XML文档分类区别于文本分类的一个重要方面,这使很多成熟的文本分类算法无法应用到XML文档分类中。因此,本文重点对XML文档的结构特征进行了研究。首先,给出一种频率一路径模型来表示XML文档的结构,该模型中不但保存了节点的标签信息,同时统计了相同路径出现的频率,使得在保证不丢失有效信息的前提下大大减小了原来路径模型的规模。其次,在频率.路径模型的基础上,给出一种带位置权重的基于路径的XML文档结构相似度计算方法WLCS(Weighted Longest Common Subsequence)。该方法在路径匹配时,使用最长公共子序列方法,能够捕捉到现有路径匹配方法漏掉的有效信息;在进行路径相似度计算时,引入位置权重向量,将路径节点的位置信息考虑在内。通过在真实数据集上做实验表明WLCS方法召回率和准确率均高于当前存在的基于路径计算相似度的方法。再次,基于频率.路径模型给出一种新的XML文档结构向量化方法。该方法使用一种基于路径频率的信息增益方法选择特征路径,使用WLCS方法中的路径相似度计算方法生成特征路径向量,并通过实验说明该方法的有效性。最后,在本文研究基础上,结合大连市公安局“全文搜索系统”的实际课题,给出XML文档分类的一个具体应用。更多还原

【Abstract】 XML (extensible Markup Language), as a common data exchanging and transmitting standard, contains rich information. Data mining on XML has become a new research focus of web mining.This article focuses on classification methods of XML document. The structure characteristics are very important features which text documents do not have, therefore most technologies and algorithms used in text mining are not suitable to XML mining because of the structure characteristics of XML documents. So, this article pays more attention to the structure characteristics of XML document. First, a model called Frecquency-Path model is proposed to express XML documents. This model not only preserves labels of correspondent nodes, but also provides frequency of same paths, so it can decrease the tree path model scale consumedly on the condition of not losing meaningful information. Second, on the basis of Frecquency-Path model, a similarity calculation method called WLCS (Weighted Longest Common Subsequence) is proposed. The longest common subsequence method is introduced for matching paths; position weight vectors which keep the position of nodes in mind are introduced for calculating the similarity of paths. Experiment results on true data set demonstrated the better recall ratio and accuracy than exsited methods. Third, a new vectorization method of the structure of XML document is proposed on the basis of Frecquency-Path model. When processing vectorization, an improved IG algorithm which is based on path frequency is introduced combining with WLCS. Finally, the research of the XML document classification method is applied in full-text searching system from Da Lian public security bureau.更多还原