An Intelligent Control Strategy for Computer Consultation.
Abstract
The heart of a computerized expert system such as MYCIN or PROSPECTOR, is embodied in an inference network of associated propositions. Inference networks are organized with rules that direct the propagation of information. A top proposition under consideration is updated by depth-first traversal of the entire networks in many current systems. We have applied the concept of 'merit' to consultant systems to direct this network traversal. Merit values force a limited best-first traversal of the inference network, requesting only the most pertinent ifnormation from the user. The notion of merit previously used in the MULTIPLE system has been extended to the MYCIN style of propagation and to the subjective Bayesian updating used in PROSPECTOR. In addition, we propose a general mechanism for merit calculation with any differentiable updating scheme. Merits will provide an intelligent, efficient mechanism for controlling new, more versatile expert consultant systems. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 08, 1982
- Accession Number
- ADA113388
Entities
People
- E. Halpern
- Jonathan E. Slagle
Organizations
- United States Naval Research Laboratory