網頁瀏覽者行為之泛化分群分析

Abstract

網路使用者行為探勘，是將資料探勘的技術，使用在分析網路使用者的行為上面。透過分群分析的技術，可以將相似的使用者行為聚集在同一群，藉以分析此行為之特性。但是由於現今的網站往往有成千上萬的網頁，所以會造成分群分析時沒有效率。為了解決這個問題，我們提出了結合網頁的概念式階層架構，並使用階層式特徵選取技術的泛化分群系統。分析者可以透過此系統，來選擇不同泛化階層的行為特徵，去做分群分析。為了能有效率的實做階層式特徵選取功能，我們提出了階層關係內嵌式索引技術，將網頁的概念式階層關係編譯到索引的編碼中，透過索引的編碼即可知道泛化特徵間的階層關係。最後透過我們的實驗，顯示了透過階層式特徵選取的泛化分群功能，可以幫助分析者得到更有意義的使用者行為族群。

Web Usage Mining is the process of applying data mining technique to web data in order to discover the access patterns of web users. Cluster analysis of the Web Usage Mining can group the users’ behaviors into clusters which have common characteristics. However, the huge amount of web pages may cause some inefficiency issues. To solve the problems, a Generalized Clustering System with Hierarchical Feature Selection Technique based on the given web pages concept hierarchy is proposed. In the system, the analyst can select the appropriate levels of generalized features to apply clustering analysis. A Hierarchy Embedded Indexing Technique is proposed to enhance the Hierarchical Feature Selection mechanism. It encodes the concept hierarchy relations into the index codes. Our experiments also show that with the Hierarchical Feature Selection Technique in generalized clustering process can help the analyst obtain more meaningful characteristics of users’ behavior groups.