AI-enabled Criteria Extraction for Multi-Criteria Decision Analysis Using Large Language Models

In decision science, when multiple alternative solutions exist, Multi-Criteria Decision-Making (MCDM) provides decision-makers with systematic methods to select the optimal solution. Implementing MCDM requires evaluating various criteria to choose the best alternative. A key challenge in MCDM is identifying the principal criteria for evaluating alternatives, which typically necessitates an in-depth Systematic Literature Review (SLR), expert knowledge, brainstorming sessions, or a combination of these methods. This article proposes a solution for criteria identification through SLR. Reviewing literature manually to determine the most suitable criteria is time-consuming, as decision criteria are rarely mentioned in abstracts, requiring full-article reviews. As a result, manually extracting criteria from a large volume of articles is both labor-intensive and sometimes inefficient. This study proposes leveraging Large Language Models (LLMs) including Llama Cloud, RAG Ollama and Gemini 2.5. The study aims at automate criteria extraction from peer-reviewed articles, enhancing efficiency, accuracy, and scalability in decision science. The proposed solution is tested in the energy supply chain, a critical infrastructure domain where decisionmaking is increasingly complex. The results of 3 LLM models are compared with traditional manual SLR results to assess capability, reliability and effectiveness. Our findings highlight LLMs as a novel tool in decision science and operations management for criteria identification, introducing an AI-enabled Decision Support System. The main contributions of this approach are: 1) Enhancing accuracy by systematically identifying decision criteria, 2) Handling multidisciplinary complexity by extracting insights across diverse documents, 3) Saving time and resources by reducing the need for full-article manual reviews. Future research will extend this methodology to other critical domains. We aim to develop a standardized criteria catalog for Critical Infrastructures (CIs) using this technology, create a smart data model, and contribute to global collaborative initiatives led by FIWARE Foundation, TM Forum, IUDX, and OASC. Our next step will focus on prompt engineering for optimized criteria selection.