DocBase - A Database Environment for Structured Documents

			   Arijit Sengupta

				Abstract


Standard Generalized Markup Language (SGML) has been universally
accepted as a standard for universal document representation. The
strength of SGML lies in the fact that it embeds logical structural
information in documents while preserving a human-readable
form. Although the original purpose of this embedded information was
to facilitate interchange and layout, the structural information in
SGML documents opens the door for processing these documents using
database techniques. SGML facilitates this goal in a twofold
manner. First, it provides a conceptual modeling tool for collections
of documents using a document type definition (DTD). The DTD serves as
a schema to which document instances need to conform. Second, the
logical structure of the documents allows query processing beyond the
classic keyword-based searches of traditional IR systems. For
instance, instead of searching for specific keywords in the whole
document, one can search for keywords in the footnotes of specific
sections.

This dissertation uses these observations about SGML as the design
principles for developing and implementing a structured document
database system. The key difference of our approach from other similar
approaches is that the design and implementation remain entirely
within the context of the SGML framework. We achieve this by using
SGML as the modeling tool of the database instances, generating SGML
documents as outputs of the queries, and also using SGML for
expressing queries.

DocBase is a prototype research system that implements most of the
concepts presented here. DocBase uses a three-level design (external -
conceptual - physical) similar to the design of standard database
systems, although the current implementation stresses on the
conceptual and physical components of the system. At the conceptual
level, we use SGML itself as the model for structured document
databases, with the database schema represented using a DTD and SGML
documents as instances of this schema. We also use an extended form of
relational calculus to handle the complex structure and path
expressions, and equivalent SQL-like and visual query languages for
posing queries. At the physical level, the data is stored in the
secondary storage controlled by a storage manager, with index
structures built on top of the documents. We use access methods over
these data structures to efficiently implement the queries.

Recognizing the importance of users in the design of systems for
document retrieval, we propose a visual query formulation method that
uses the principle of familiarity to make the querying process easier
and more satisfying for users. We show that even at the simplest
level, this method is no less efficient or accurate than the
traditional form-based query formulation, but is significantly more
satisfying.  In addition, we show that this method is equivalent to
the other query languages proposed in this dissertation.