以本體論為基礎之圖形分割式知識融合

Abstract

對於有許多不同知識來源，且要求快速反應的工作來說，知識融合是一個很適當的選擇。以往由資料探勘中的概念性分群法所演進而來的一些方法，並不足以處理現今較複雜的知識表示需求。近來也有一些關於這個問題的相關研究，但仍然沒有一個實際上可用的架構。在這篇論文中，我們特別針對後設知識的建構，提出知識重建的概念以及三階段的知識融合架構。在此架構中，我們也定義了一種中介的知識表示法──知識關係圖，以及兩個關於知識融合過程的原則。我們也提供了關於此架構的數個演算法。最後，本論文亦將此架構應用在一個關於網路入侵偵測系統領域，進行知識融合的實作測試。

For a variety of knowledge sources and time-critical tasks, knowledge fusion seems to be a proper concern. Previous researches evolved from the conceptual clustering approaches of data mining are not capable to deal with complex knowledge representations. Recently, some researches focus on the problem, but there is still no practical framework. In this thesis, we propose a reconstruction concept and a three-phase knowledge fusion framework which addresses the problem of meta-knowledge construction. In the framework, we also define relationship graph, an intermediate knowledge representation, and two criteria for the fusion process. Algorithms of the framework are also provided. An evaluation of the implementation of our propose knowledge fusion framework in the intrusion detection systems domain is also given.