DESIGNING OBJECT-ORIENTED REPRESENTATIONS FOR REASONING FROM FIRST-PRINCIPLES

Abstract

Modeling expert knowledge using "situation-action" rules is not always feasible in knowledge intensive domains involving volatile knowledge (e.g., trading). The explosive search space involved in such domains and its dynamic nature make it extremely difficult to setup a rule base and keep it accurate. An alternative approach suggests that in some domains many of the rules expert use can be derived by reasoning from "first-principles". That approach entails modeling experts' deep knowledge, and emulating reasoning processes with deep knowledge that allow experts to derive many of the rules they use and justify them. This paper discusses the design and implementation of an object-oriented representation for the deep knowledge traders utilize in a business domain called hedging, which is knowledge intensive and involves volatile knowledge. It illustrates how deep knowledge modeled using that representation is used to support reasoning from first-principles. The paper also analyzes features of that representation that we have found to be extremely beneficial in the development of a knowledge-based system called INTELLIGENT-HEDGER. Based on our experience we feel that, with minor modifications, this representation can be used in other managerial domains involving financial reasoning.