Exploring Urban Place Function through User-Generated Textual Content

Abstract. Understanding the functions played by different places as part of the urban fabric is essential for the sustainable development of cities. The increase in the amount and variety of people's digital footprint, combined with the recent emergence of Large Language Models (LLMs), holds the potential to derive novel and useful insights into the function of urban places from unstructured textual content. This paper presents a reproducible and robust methodological framework to extract place functional characteristics from user-generated textual content. In particular, we analyse tags of Points of Interest (POIs) from OpenStreetMap (OSM) and summaries of geo-tagged articles from Wikipedia through spatial clustering and topic modelling, exploring classic and current approaches in Natural Language Processing (NLP). We evaluate and compare the performance of different LLMs, including Mistral and Llama. Our results show that the framework provides a good insight into place functionality for our case study.