網頁文獻叢集化搜尋

Abstract

隨著資訊時代的來臨，數位化文獻的資料量也以急劇速度不斷地成長。如何能在大量的數位化資料中迅速尋找出高關聯度的資料、淬取出相關的知識，無疑是一個急迫解決的重要議題。我們在這篇論文中提出一個叢集化的方法(Literature Clustering Search, LCS)。利用這個方法可以將大量的資料分類成階層式叢集，並進一步幫助使用者在短時間內能對大量的資料進行初步的了解以及具有初步的概念。我們的方法共有四個步驟。首先，Metadata Retrieval可以將資料做格式化的動作。第二步，進行Feature Selection的程序，在這個步驟中我們只留下具有文章代表性的單字或單詞做為Feature。第三步，再利用Association Rule Mining的程序計算出所有Feature之間的關係。最後，我們依據這些關係形成一個階層式叢集。由於Association Rules代表著一群共同出現的字詞，我們可以藉由這一群共同出現的字詞輕易地了解群組中所代表的涵意。除此之外，我們同時建立了一個線上文獻叢集化搜尋服務，以展示我們的方法與成果。

In the past two decades it has been seen a dramatic increase in the amount of information or data being stored in electronic format. Retrieving relevant information from large data set becomes important issue. We propose a clustering method which generates hierarchical clusters and helps us to have overall picture of the concepts through the massive information in a short time. We call it Literature Clustering Search (LCS). There are four steps to accomplish the task. First, metadata retrieval will help normalizing the data format. Second, feature selection extracts words/phrases which could represent the document. Third, association rule mining generates relations between features. Finally, group the documents that share the same association rules. Since association rules represent a set of terms that co-occur frequently, we could easily obtain the concept of the cluster based on the association rules of the cluster. In addition, we build an online clustering web service to demonstrate the methodology of literature clustering search.