Transcript Why OODBMS
資料庫與資訊檢索系統的整合 - 一個文件資料庫系統的開發研究 by Chung-Hong Lee (李俊宏) Assistant Professor Dept. of Information Management Chang Jung Christian University AGENDA • Introduction • Comparison of the DBMS and IR approaches for document retrieval • Proposed signature based IR technique • System architecture • Integration method • Conclusions A Document Retrieval Model Document collections User query Query formulation Indexing Signature of query Signature of documents Matching of similarity Relevance feedback Results of retrieval Convergence of Information Retrieval and Database Information Retrieval Database proprietary application dependent rich search capability standardization application independent limited search capability standardization application independent powerful modeling capability powerful search capability rich application development tools Related work • A text-search extension to the ORION OODBMS developed by Lee (1991). • The integration of the INQUERY text retrieval system and the IRIS OODBMS proposed by Croft (1992). • Mapping the SGML document structures into OODBMS’s data models: –Christophides (1994). –Macleod (1995). –Volz (1996), etc. Differing from some of the above efforts with the aims to model only SGML documents in DBMS, our system is particularly aimed at handling heterogeneous types of documents, such as textual and multimedia documents, and providing content-based retrieval functions to describe the stored document objects. Why OODBMS ? The core features of OODBMS supported by most such systems are: 1. Complex objects 2. Object identity 3. Encapsulation 4. Types and Classes 5. Class or Type Inheritance 6. Overriding, overloading and late binding 7. Computational Completeness Signature file approach (I). Signature Generation: word Object Database Management System Coding 0010 0000 0000 0000 1000 1001 0000 0000 …… …… …… …… result 0000 0000 0001 0000 0001 0000 1001 0001 OR S(P) 0010 1001 …… 0001 1001 (document signature) . (II). Signature Matching: Signature of a query AND S(P) Result 0000 0000 …… 0010 1001 …… 0000 1001 0001 1001 0000 0000 …… 0000 1001 Concept of the scalable signature file approach • document signatures are generated according to their composed Chinese characters • the document signatures are divided into two segments: the first segment represents the occurrence of commonly-used Chinese characters, while the second segment represents the occurrence of the remaining Chinese characters and the English character bigrams • the signature size can be adjusted with the average length of each document System Architecture (1) Document indexing module OODBMS GUI wrapper Text Retrieval Engine retrieved interface document objects for full-text search document input System Architecture (2) Key features:- OODBMS Two stage search Both IR and OQL queries are available Signature file as a preprocessor for IR queries Documents are stored as BLOB object representation in the OODBMS Search Engine Signature file OODB Search engine IR queries (word, term & phrase) OQL queries retrieved document objects Queries Signature file as a pre-processor of the database queries (path expression) Information filter Signature file list file L (list of candidate objects) #1 ……. #2 ……. #3 ……. #4 ……. #1 0010 0010 0000 0001 0010 1000 document.plaintext_doc.engineering.id5 #2 0010 0010 0000 0000 0010 1000 document.hypertext_doc.SGML.id117 #3 0000 0010 0000 0001 0000 1000 document.plaintext_doc.engineering.id7 #4 0010 0000 0000 0001 0010 0000 document.compound_doc.voice_mail.id8 ……………. OQL queries to OODBMS Query text processing How the system formulates the query:The system transforms Quasi-Natural language queries incrementally into complex structured queries in the query language. Goal: Free format queries Related techniques:• • • • • Key term extraction from the queries IR-queries-to-OQL conversion Query optimization User interface NLP Conclusions The distinctive features of underlying system developed :• IR-OODBMS integration – OODBMS based document repository – a loose coupling approach – signature file filter as a preprocessor for query processing – two stage search – a novel query model – easy to maintain, including the signature file and database schema • Signature generation – a character based signature method designed for Chinese and English documents • Applicable to a digital library infrastructure