A Graph-based Similarity Function for CBDT: Acquiring and Using New Information

Abstract

We present a formal model of decision-making under uncertainty, a variant of Case-Based Decision Theory, in which the solution to a problem obtains in terms of the distance to previous problems. We formalize this by defining a space based on an orthogonal basis of {em features} of problems. New problems are evaluated in this setting, determining the best available actions in terms of the distance to previous problems. This is particularly relevant in the case of Accounting, in which information systems have to support efficient ways of solving problems. We also show how this framework evolves upon the acquisition of new information, namely features or values of them arising in new problems. We discuss how this can be useful to evaluate decisions based on not yet existing data.