Browse Theses

Expert system knowledge bases have traditionally been manually loaded by a knowledge engineer. The "knowledge" was first extracted from a expert through a series of questions and answers conducted by the knowledge engineer. Then through establishing rules and/or specific examples from the real world, this information was carefully coded and loaded into a knowledge base. This is a process that sometimes took years to complete.

More recently, advances in expert system technology have addressed reducing the involvement of the knowledge engineer and providing tools for users/experts to build their own knowledge bases. Most of these expert system building tools rely heavily on the human expert not only to provide the knowledge, but also to provide the logic structure of the knowledge. The next step in the enhancement of expert system development tools is not only to reduce the time required in collecting knowledge, but additionally to structure that knowledge into a usable expert system shell format. Rather than develop "intelligent interface" techniques through direct dialogue with a "human" expert, this study addresses the development of automatic knowledge base acquisition techniques that use "text" as a source of knowledge.

Hard copy "text" knowledge is readily available in many areas suitable for expert system development. Acquisition of knowledge from text has not been successful thus far, primarily because most text is not presented in a format (rules, frames, or logic) that can be directly used to load a knowledge base. Most text books, manuals, training guides, etc. are written in a conceptual presentation format. This format may be amenable to human comprehension, but it generally does not spell out the "what-if-else" type details required to construct an expert system knowledge base. This study proposes techniques to over-come these inherent problems.

This study develops a model to illustrate the techniques of automatic knowledge base acquisition from text. Several examples are used to show the theory of the techniques. A real world application is selected, and two validation exercises are conducted to support the credibility of the model. This study identifies and illustrates new automatic acquisition techniques that may greatly enhance the speed and structure with which knowledge bases can be constructed.