Article ID: | iaor19971475 |
Country: | Netherlands |
Volume: | 65 |
Issue: | 1 |
Start Page Number: | 181 |
End Page Number: | 193 |
Publication Date: | Aug 1996 |
Journal: | Annals of Operations Research |
Authors: | OLeary Daniel E., Schaffer Joanne |
Keywords: | programming: linear, programming: dynamic, programming: integer, programming: goal |
The purpose of this paper is to use mathematical programming, including linear programming, dynamic programming, integer programming and goal programming to verify rule-based knowledge. The authors investigate both domain independent verification, exploiting the general structure of rules, and domain dependent verification, exploiting structure in the domain. Mathematical programming software is readily available and is very efficient. As a result, verification using mathematical programming can be very efficient at finding errors. Mathematical programming can be used to more than just find errors in knowledge representation. Once an error has been found, mathematical programming can be used to ‘recommend’ an alternative. The recommendation can take into account the previous verified knowledge to mitigate the potential introduction of redundant knowledge and to help guide the choice process. Normally the development of recommendations to fix errors has been ignored in the verification literature, and treated as a separate knowledge acquisition task. Accordingly, this paper also extends the verification effort by providing a recommendation on how to fix errors.