LLM + DEX: Faster and Explainable Construction of Hierarchical Decision-Making Models

Abstract

This paper explores the possibilities of integrating Large Language Models into the DEX methodology for multi-criteria decision-making. DEX models are interpretable, hierarchically organized decision systems based on qualitative rules and preferential logic, but their construction requires time and domain expertise. On the other hand, LLMs, such as ChatGPT, possess the ability to process text and generate decision structures based on problem descriptions and data. In this paper, we propose a method that uses LLMs to select relevant attributes, construct a hierarchy of aggregate criteria, and generate semantic explanations, while decision rules are learned from data with monotonicity constraints. On an experimental case of ski injury risk modeling, we demonstrate that the resulting models achieve performance comparable to the DIDEX model and standard machine learning models, with significantly faster development and improved interpretability. The proposed approach opens opportunities for further automation and the development of intelligent agents for decision-making support.

Sandro Radovanović

Assistant Professor at University of Belgrade

My research interests include machine learning, development and design of decision support systems, decision theory, and fairness and justice concepts in algorithmic decision making.