Integration of Large Language Models (LLMs) in Multicriteria Decision Support Methods: A Systematic Review of the Literature

Large Language Models (LLMs) tools were widely used after the popularization of OpenAI's ChatGPT and began to be applied in various areas of knowledge. From the perspective of multicriteria decision models, there is potential in the use of these tools in the stages of structuring, elicitation and evaluation of preferences, as well as in the role of artificial decision-makers. However, this application between multicriteria methods and LLMs is still incipient in the literature, which characterizes a relevant gap. Therefore, this work answered two questions: which approaches have been used to combine LLMs with multicriteria methods and which model can guide this use. The objective is, based on a systematic review of the literature, to categorize approaches and propose a utilization model based on the multicriteria framework. The review was conducted on the Scopus, ResearchRabbit, and Google Scholar platforms, with a focus on articles published in journals after 2017, the year of the seminal article on transformers. 5,807 records were identified. As a first filter, a Python script was developed that enriched the database from the abstracts and signaled the mention of multicriteria methods (ELECTRE and PROMETHEE FAMILIES, MACBETH, AHP, ANP, SMART, SMARTER, Tradeoff, FITradeoff, DEMATEL, TOPSIS and VIKOR), leaving 98 articles. In the reading of the abstracts, paired classification was adopted between one of the authors and the Claude and Gemini tools, allowing double verification; In the end, 31 articles were read in full and 24 were considered relevant and synthesized