二項圖中伴隨不穩定性資料的單二元分類

Abstract

隨著資料相關性的產生與其概念的崛起，傳統的分類方法在如今多變的資訊 形態中已經受到了挑戰。單單只依靠像是屬性等對於資料而言是獨立的資訊已不 足以推導出實際的資料分類架構。為了解決這個問題，研究人員們發展出一系列 稱為Collective Classification 的分類方法。這類的方法利用了未觀察資料［或可稱為未分類資料〕與其彼此之間相關的特性來幫助分類，並確實的提升了分類的準確度。然而由於資料收集的不完全或是其他因素所造成的資料不穩定性則讓這類方法在這種情況下分類的時候因為資料的不穩定而無法得到足夠好的分類結果。在這篇論文中，我們專注於研究如何在伴隨不穩定性資料的情況下來對未觀察的資料做單二元類別的分類。我們同時也在此提出了GLRD［群體類別比例分布〕一個嶄新的概念。藉由計算不同GLRD 的相似度，我們可以推導出有著高 準確率的分類結果。

With the rise of the relevant data generation and its concept, the traditional classification method has been challenged in changing information form today. Just relaying on the independent information of itself like attributes is not sufficient to derive the actual data classification structure. To solve the problem, researchers have developed a series of classification called Collective Classification which takes advantage of unobserved relevant data information to classify. However, data uncertainty caused by data collection or other factors makes these collective classification approaches not get good enough results in some cases. In this paper, we studied the problem of inferring single binary class label of the unobserved data with uncertainty and proposed a novel concept, called GLRD(Group Label Ratio Distribution). By calculating the difference of GLRD’s similarity, we can derive the classification with high accuracy.