基於規則庫切割的元知識建造方法

Abstract

在資訊科學的範疇中，專家系統（Expert System）的應用越來越為廣泛，儼然成為下一代資訊系統的重要特色；而在專家系統的建置技術方面，規則庫（Rule Base）是一種廣為採用的方式，在規則庫中的知識，被設計為人類容易瞭解的邏輯式規則，更使得規則庫無論在設計上、應用上，都能夠符合知識工程的需求。然而，由於資訊系統的進步以及電腦硬體能力的增強，使得規則庫的規模有越來越為龐大的趨勢，因而造成規則庫的規模增加，從此產生了許多管理上的議題。在本論文中，藉著結合規則庫切割（Rule Base Partitioning）和元知識（Meta-knowledge）建造機制，RP-MES被提出來解決規則庫管理議題，其中透過對於規則庫結構與語意的分析，協助建置更有效率且更容易維護的規則庫。此外，我們也設計及實做一基於RP-MES的入侵偵測雛型系統，以及關於系統效能的數個實驗。實驗結果顯示，透過RP-MES可以將規則庫分割為合適大小的規則群，並維持知識的正確性，同時顯示最終的執行效能相較於原本不經過分割的規則庫之效能有著明顯的提昇。

Expert system technology becomes more and more important in computer science domain for next generation computer systems. For constructing an expert system, rule base is a widely used approach, where knowledge and expertise are represented as production rules. However, due to the growth of rule base usage, the scale of rule base is increasing and hence many management related issues arise. By designing a new approach combining both rule base partitioning mechanism and meta-rule construction mechanism, RP-MES is proposed to solve these issues in this thesis. An Intrusion Detection System (IDS) prototype is also designed and implemented based on RP-MES, and some experiments have been done to evaluate the system performance. The experimental results show that RP-MES can produce reasonable number of rule clusters and the accuracy of the inference result remains, and that the performance of RP-MES is better than that of original rule base without partitioning.