一個發現變異物件的知識擷取方法

Abstract

隨著知識庫系統的演進，也伴隨著產生許多的變異物件，而傳統的知識擷取表格，無法有效的發現變異物件。且沒有相關的研究指出，知識庫能夠隨著推論結果的演化。在本論文中，我們提出一個發現變異物件的知識擷取方法，找出隱藏在實際世界中的變異物件。由紀錄那些經常被推論且有著較低信賴指數的隱藏規則，並藉由新增物件，及兩個額外的運作: 改變屬性的型態及新增加屬性。這些運作可以幫助專家來找尋變異物件及提高隱含規則的信賴指數(CF)。為了實現我們的想法，我們亦提出一個EMCUD-DRAMA-VODKA的架構。在這個架構底下包含了知識產生，知識利用，及知識發掘三個階段。首先我們運用 EMCUD來產生原本和隱含的規則。在知識利用階段，我們運用DRAMA所產生實際的運用上的使用紀錄，最後在知識發掘階段，我們運用VODKA 來發現變異物件及屬性。我們也完成包含三個階段的整個系統的實作。並且我們用蠕蟲來做為我們的例子。

The traditional knowledge acquisition process might be useless for solving the problems of variant object discovery. It might never be mentioned for the knowledge base to evolve with inference results. So the purpose of this thesis is to present a Variant Object Discovering Knowledge Acquisition (VODKA) to add a new object from the inference results, including two operations: changing attribute data type and adding a new attribute. These operations assist experts in discovering the variant objects and elevating the Certainty Factor (CF) of embedded rules. We also propose the framework of EMCUD-DRAMA-VODKA which has three phases: Knowledge Generation, Knowledge Utilization, and Knowledge Discovery. To begin with, we use Embedded Meaning Capturing and Uncertainty Deciding (EMCUD) to generate the original and embedded rules. Moreover, we apply these rules in the real world and accumulate the logs of DRAMA in the Knowledge Utilization phase. Finally, the VODKA is proposed to discover the variants or attributes in the Knowledge Discovery phase. In addition, a worm taxonomy classification example is given. The implementation of these phases is also constructed for describing how to develop a complete knowledge-based application.