分群屬於資料挖掘中一個重要的領域，但在以往的研究中，分群方法大多無法讓分群結果同時具有語意描述，其中CLTree雖然可以克服這項缺點，但仍有一些弱點存在，也就是用來建立語意描述的屬性與分群的屬性是完全相同的，實務上卻有可能遇到用來建立語意描述的屬性與分群的屬性不完全相同、甚至完全不同的情況。例如網站設計者在重新設計網站架構時，常一起被存取的網頁表示彼此間有相似的特性，便可將一起被存取的網頁記錄視為分群屬性對網頁做分群；但用以建立網站架構的分類目錄的屬性不可能也使用一起被存取的網頁記錄，因為這種目錄不易於了解也不具有語意上的任何意義，我們通常會另外選擇有意義的屬性來做分類，例如網頁內容的主題、關鍵字、撰寫時間等，而在本研究中語意描述的屬性也就是所謂的分類屬性。 因此在本研究裡，擴展CLTree的概念，發展出三種可以處理分類屬性與分群屬性不需完全相同的具語意描述能力的分群演算法。 In the previous works, most of the clustering methods can't give each of the clusters a semantic description. There is a novel clustering method, CLTree, which can solve such a problem. However, the attributes which are used in clustering are the same as the ones used in establishing the semantic descriptions. We may meet a situation that the attributes used in clustering are different from the ones in establishing semantic descriptions. For example, when redesigning the architecture of a web site, we may use the browsing log to be the clustering attributes since the pages which are often access together imply that they have more similar properties. At the same time, we would choose other attributes, such as subjects, keywords, or last modified time of web pages, to build the hierarchical directory because browsing log is meaningless in interpreting the website's architecture. Note that we use classification attributes denote attributes which can used in establishing semantic descriptions. In this paper, we extend the concept of CLTree and develop three clustering algorithms with ability of semantic description. These algorithms can use different classification attributes and clustering attributes.
