Summary: | 碩士 === 國立臺灣科技大學 === 工程技術研究所 === 81 === Integration of expert systems and database systems promises to
make expert system work efficiently when it needs a very large
knowledge base, while endowing database systems with
inferencing capability. We designed and implemented a system,
called TIGER, that is a total integration of expert systems and
database systems. In our system, all the facts of in
traditional expert systems are stored in the relational
database component of TIGER. All partial matchs, intermediate
facts, and rule instantiations are also stored in the temporary
database. To execute the rules of expert systems, we chose the
TREAT match algorithm, appropriately modified, as a basis for
rule matching because of its superior efficiency in comparison
with other match algorithms and its suitability for integration
with a DBMS. However, traditional rule execution strategy is
sequential, resulting in extremely high frequency of disk
access, and low overall system performance. We propose a novel
and vital idea in our system to siginifcantly reduce the number
of joins by applying parallel rule firing theory to rule
execution in expert database systems. The intuition behind the
idea is that when multiple rules are discovered to be firable
in parallel, their joins can be performed at the same time,
instead of being done separately, as traditional sequential
execution of rules demands. Our experimental results confirms
our intuition and demonstrates that as more parallel rules are
present in the application, and as the data set gets larger,
the system achieves very impresessive performance.
|