PersonaCraft: Leveraging language models for data-driven persona development

Abstract

Generative AI, with its large language models (LLMs), provides various opportunities for the development of user-centric systems in human–computer interaction (HCI). Yet, use cases of LLMs in HCI are still scarce, calling for developing and evaluating real systems. We present PersonaCraft, a data-driven persona system using LLMs to address this need. The system analyzes a common source of user data – surveys – and generates personas, humanized representations of real segments in the data. By integrating LLMs with survey data analysis, PersonaCraft combines persona development and modern artificial intelligence methodologies to provide researchers and designers with user-centric insights from nearly any survey dataset about people. Various evaluations of the system, including with internal evaluators, general users (n = 127), and user experience professionals (n = 21), indicated that PersonaCraft personas scored high on all evaluation criteria of clarity, completeness, fluency, consistency, and credibility. The application of PersonaCraft can extend across a range of domains, including user research and population-level people research.