 Abstract | The introduction of databases for data storage and handling revolutionised the
way we dealt with records and enabled simple and fast information processing, aggregation
and summarisation. Database and information technology systems have
evolved from simple file processing systems to powerful database systems. Data
management technology has progressed from hierarchical and network systems
to relational databases, data modelling tools and indexing and organisational
techniques. The development of Relational Database Management Systems and
automated systems put the layout and form into the unchanging metadata and
gave us record once systems.
Unfortunately, the 'real world' upon which databases are modelled constantly
changes. These changes may affect the schema for a variety of reasons including;
- Unanticipated requirements,
- A change in the universe of discourse,
- A change to the interpretation of facts about the universe of discourse,
- Changes in the form of updates to effect upgrades to the functionality or
scope of a system,
- Changes in the form of updates to effect efficiency improvements,
- Changes caused by system operation,
- Error correction.
Different formalisms have been developed to deal with schema changes with
the aim being to preserve information capacity and preserve semantic correctness.
Schematic changes may be the result of evolving one system or may arise due to
the need for merging two or more systems. Schematic conflicts occur which
must be resolved and the schemata unified to produce a new version. To reach
this goal there are graph based schema integration architectures, as well as, semiautomatic
systems applying schema matching and schema translation techniques.
These systems also utilise ontologies, thesauri, and so forth to integrate data from
heterogeneous sources in order to process queries and views.
Data integration or conversion remains a partially resolved issue. Some metadata
changes are managed by changes to application code and system down time
for conversion procedures. However an attribute change may result in data loss,
changed accuracy, and altered semantics. Whilst the use of ontologies, concept
graphs and other knowledge interchange techniques are alleviating the problems
of data integration, these structures are not yet an integral part of the database
architecture.
This thesis argues a three-level architecture for relational databases with an
interface positioned between data and metadata for complex domains. This intermediary
level is the mesodata layer. This mesodata layer, separate from the
metadata and data, provides complex structures, such as graphs, queues, and
circular lists, in which to store domain values and their inter-relationships as well
as supplying the 'intelligence' required to operate and manipulate them. The
domain structures enable different orderings that form the bases of filters for
enhanced querying and information retrieval. DBMS supplied mesodata types
would allow for the re-usable inclusion of domain information such as in ontologies,
taxonomies and concept graphs that to date have been only application
specific. |
