Large Language Model–Assisted Discovery of Psychological Constructs: Method and Application

Abstract: Large language models offer new possibilities for proposing psychological constructs, but the distinctiveness and research value of the concepts they generate require systematic comparison and evaluation. This study presents a workflow for generating and screening candidate psychological constructs with large language models. Candidates are screened for semantic similarity against existing concepts and previously retained candidates. Model-generated feedback guides revisions to preliminary definitions and explanations of conceptual differences, and researchers independently assess the candidates’ research value. In an application to life orientations in a society undergoing transformation, the selected candidates concerned purposes and responsibility boundaries in interpersonal relationships, kinship-like relationships with virtual characters, and ways of deriving satisfaction from different life domains. One selected candidate was further examined through the analysis of naturally occurring texts to clarify its meaning and develop criteria for identifying relevant expressions. By incorporating conceptual comparison and expert evaluation into model-assisted exploration, this study offers a practical approach to proposing and prioritizing candidate psychological constructs.