Knowledge acquisition and refinement in expert systems
Abstract
The issue of knowledge refinement in expert systems is addressed in this thesis. In general, an expert system is composed of a knowledge base which stores application specific reasoning knowledge, an inference engine which processes the stored knowledge, and an interface through which communication links between the users and the expert system are established. In terms of knowledge refinement, this architecture is dependent on knowledge engineers to refine its stored knowledge on a periodical basis. The frequency with which the knowledge base is revised depends very much on the underlying application domain. Furthermore, the control mechanism of the inference engine may also need to be updated in order to match up with the changing inference process of human experts. In the scope of this thesis, we will primarily focus on the former. In this thesis, we will generalize the principle of knowledge acquisition to knowledge refinement of a continuous nature. While knowledge acquisition takes place in the early stage of an expert system development, knowledge refinement is applicable during the entire life-span of an expert system. The thesis starts by presenting a conceptual framework of building knowledge acquisition systems. Based on this framework, a generic architecture of expert systems with a provision to refine its own knowledge base is discussed. The novelty of this research is to study knowledge acquisition and refinement in expert system by presenting an architecture and to prove its validity, at least partially, by streamlining it to some generic problem tasks which are illustrated with a real-life application. The thesis is organized as follows: Chapter 1 presents a conceptual framework of building knowledge acquisition systems. The idea of knowledge acquisition is extended to knowledge refinement in Chapter 2; a generic architecture of expert system that are capable of self-refinement of knowledge is presented. The architecture is then used to construct expert systems to perform generic problem tasks of pattern recognition and classification in Chapter 3 and Chapter 4 respectively. Chapter 5 concludes the thesis by discussing future research directions of knowledge refinement.
Degree
Ph.D.
Advisors
Whinston, Purdue University.
Subject Area
Management
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.