基于Dirichlet过程的Deep Web数据源聚类方法.pdf

Web数据源聚类方法 基于Dirichlet过程的Deep 黄进，何中市，李英豪 (重庆大学计算机学院，重庆400044) 摘 要：提出了一种基于Dirichlet过程的Deep 过程(HDP)进行特征提取。首先将查询接口中原本高维稀疏的文本表示为主题特征，该过程能自动确 Web数据源数量大、变化快的特 聚类个数，由Dirichlet过程根据数据自动计算得到。特别适用于Deep 比，均取得较好的结果。 Web；数据集成；特征提取；dirichlet过程；混合模型 关键词：Deep 中图分类号：TP39l 文献标识码：A of web onDirichlet Clusteringdeep sourcesbased process Jin，He Huang Zhongshi，LiYinghao of (College 400044，China) Computer，ChongqingUniversity，Chongqing Abstract：This a methodof websourcesbasedondirichlet method paperproposedclustering deep process．Theproposed thehierarchicaldirichlet featureextraction．Firstit the textinthe adopted process(HDP)for replacedhigh-dimensionalsparse with off宅atures．Thenintroducedthedirichlet interface featureandit could determinethenumber searching topic automatically mixturemodelforthetext thetextasa muhinomialmodel．Basedonthedirichlet model process clusteringbytreating process，this could determinethenumberofclusterswithoutmanualintervention．Itwas suitableforthe automatically any especially of web is of scaleand withother results sources，which deep large changesfast．Compared clustering TEL-8 demonstratea onthedatesetsof onF—measureand goodperformance Entropy． web；data extraction；diri