資訊爆炸時代的來臨，越來越多使用者在網路上搜尋相關資料進行閱讀。本研究目標是將大量文件資料進行階層分群（Hierarchical Clustering），並以字詞關係建置具有上下包含關係的分類學（Taxonomy），以用來成為階層群集的標籤。運用上，能方便使用者快速瞭解文件集有哪些主題，迅速選擇所需主題的文件進行閱讀。本研究提出的系統架構有效地改善了階層群集研究上的五個議題：高維度的向量、動態的特徵選取與文件分群、文件處理順序、文件跨領域分群與群集標籤之間的關係。 With the popularity of Internet, the World Wide Web contains a giant amount of information. To search relevant information from large number of texts becomes a challenge to the users. Hierarchical clustering is one of the methods to conquer this problem. Because its features let users can browse the topic gradually and find out the most relevant documents they have interesting. But there are still have some challenge in hierarchical clustering must be addressed, like high dimensionality of the data, dynamic data sets, the sensitivity of input order, documents has several concept, and the relationship of clusters and tags. In this paper, we propose an approach of dynamic hierarchical clustering based on taxonomy to conquer those challenges. The experimental result shows that our method not only suitable for constructing hierarchical clustering in dynamic data sets, but also offer a easier structure to browse than traditional algorithms, BKM and UPGMA. In addition, the clusters are labeled meaningful tags with the relationship of containment can let users understand the whole concept of clusters rapidly.
