Using Markov Decision Process for Recommendations Based on Aggregated Decision Data Models

Our research is placed in the context of business decision making processes. We look at decision making as at a workflow of (mostly mental) activities directed at choosing one decision alternative. Our goal is to direct the flow of decision activities such that the relevant alternatives are properly evaluated. It is outside our purpose to recommend which alternative should be chosen. Since business decision making is data-centric, we use a Decision Data Model (DDM). It is automatically mined from a log containing the decision maker's actions while interacting with business software. The recommendation is based on an aggregated DDM that shows what many decision makers have done in the same decision situation. In our previous work we created algorithms that seek a local optimum. In this paper we show how the recommendation based on DDM problem can be mapped to a Markov Decision Process (MDP). The aim is to use MDP to find a global optimal decision making strategy.