abstract = "The need was identified for a generic Information
Retrival tool. Genetic Programming was selected as most
suitable paradigm for providing the necessary adaptive
intelligence. This was combined with conventional
Bollean query search techniques. Each Query is treated
as a genetic individual and a population of these is
eveolved so as to move through the search space of all
possible queries efficeiently. The criteria that guide
this search is termed relevance feedback. This
information is derived from the suer through tne
evaluation of a document set and forms the basis of the
fitness funtion. The best query produced in this way
can then be used to scan other documents, ordering
these according to relevance. These processes can be
lined to produce an application that can learn by
experience, requires no explicit instructions and can
be apllied to a wide variety of IR situations.
The development work was divided into three stages:
design and implementation of an experimental software
platform, research into viable configurations using
this platform, and construction of working models.
Stage one formed the focus of this project.
The project specification was thus to produce a
software system that can act as a testbed during
experimentation in teh second stage and as an early
prototype of future applications. This was achieved,
the souce code being written in C++ to run on a PC.",