特徵檔之設計及其在物件導向式資料庫之應用

Abstract

特徵檔在過去的研究中被廣泛的應用在關聯式資料庫的文書擷取法上，近 來的研究發現特徵檔的特性可用來解決物件導向資料庫的一些問題。本論 文主要分為兩大部份，在第一部份中，由於特徵檔法的優劣取決於兩個因 素，一是錯置機率的大小，另一是搜尋特徵檔的時間，而在過去研究中都 被分開單獨考慮。因此本論文的主要目的之一就是同時考慮這兩個因素， 提出結合特徵擷取方式與特徵檔儲存結構的高效率特徵檔法。基於這個觀 念設計出新的特徵檔法，同時具有高過濾能力和快速搜尋特徵檔的優點。 第二部份則應用特徵檔法解決物件導向式資料庫中索引的問題。在過去的 研究中，物件導向式資料庫中的索引，只能針對特定的查詢建立特定的索 引，十分不便，因而有人將特徵檔應用在索引上，使索引能一般化。因此 本論文另外一個主要目的在於提出一個較前人研究成果更一般化且更省時 間的新方法。 Signature file is a popular strategy applied to the text retrieval methods of relational database. Recently, some researches discovered that the characteristic of signature file could deal with some problems of object-oriented database. There are two primary parts in this thesis. Firstly, it proposes a new method of signature file. Secondly, signature file is applied to object - oriented database. Part I : There are two factors that influence the efficiency of the signature file. One is the false drop probability, the other is the cost of searching the signature file. The false drop probability is affected by the signature extraction methods, while the cost of the signature file is decided by searching storage structures. These two factors were considered separately in the literature. In this thesis, the signature extraction methods are integrated with the storage structures to result in several new signature file methods. The new methods take the two factors into considerations at the same time. The new signature file methods designed upon this concept have the capabilities of high filtering and fast searching. Part II: In the past, the index of OODB was designed for special queries. It is inconvenient and not general. In this thesis, a new method based on the signature file approach is proposed to solve the problem of index in OODB.