Data Assisted Knowledge Acquisition for Alteration of Table-Based Expert Networks.
Abstract
Two main approaches to knowledge-based systems are expert systems and network systems. Researchers at Florida State University developed expert networks to combine the benefits of expert systems and network systems, including the explanation capability of expert systems and the learning capability of network systems. Expert networks still rely on the knowledge engineer's ability to determine the connection relationships among inputs, the set of stored facts and rules, and conclusions. The interpretation of gas chromatography data provides the application domain of this research. The Contaminant Analysis Expert Network project involves applying expert network technology to this problem domain in which analytical chemists use knowledge tables to represent the domain knowledge. The work of this thesis centers on the need to build automated systems that utilize example data to aid in knowledge acquisition and structural alteration of these table-based expert networks. The new approach to network structural alteration presented here automates the ability to confirm, refine, and augment expert knowledge. The automated tool developed in this thesis combines the knowledge derived from experts with the relationships discovered in data. In altering the expert networks to improve prediction performance, the tool maintains the ability to retrieve the expert knowledge from the network. The result is expert networks which structurally capture both the table-based knowledge of analytical chemists and information gleaned from example data. These networks outperform networks built from expert knowledge alone.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1995
- Accession Number
- ADA300722
Entities
People
- Robert G. Timpany
Organizations
- Florida State University