Abstract

The emergence of Large Language Models (LLMs) has changed the world and human lives in many different ways. Nowadays there is a new idea that LLMs are a big database of people’s opinions, therefore, It is possible to replace them with humans in some parts of the design process, while LLMs may not capture all the individual and cultural differences, and it can lead to discrimination against a group of people. This thesis investigates the representativeness of LLM-generated user personas by comparing three different persona creation approaches-LLM-solely, LLM-auto, and LLM-summarizing-across two cultural contexts: Germany and Iran. By collecting survey data from real participants in both countries, this study establishes a ground truth for evaluating the generated personas with LLM.

Keywords: Large Language Model (LLM), User personas, Human-Centered Design (HCD), Similarity score