Advancements in the application of large language models in urban studies: A systematic review

Abstract

Large language models (LLMs) possess robust comprehensive capabilities for addressing complex problems and have increasingly been applied to better elucidate urban phenomena. Amid the surge of LLMs, there is an urgent need for urban researchers to understand how previous works have integrated LLMs across various disciplines. In this paper, we conduct a systematic review of 233 articles focusing on the application of LLMs in urban studies, analyzing the developing tendencies using information extracted from articles with the aid of a customized Generative Pre-trained Transformer (GPT) assistant and giving in-depth reviews across different subfields. The findings highlight an exponential growth in related research over the past six years, with LLMs being applied to diverse scenarios such as text analysis and generation, domain knowledge question-answering as well as field-related task. Besides, GPT-based and Bidirectional-Encoder-Representations-from-Transformers-based (BERT-based) have emerged as the two mostly used models, with embedding and fine-tuning being the predominant methods for data processing and model adaptation. Additionally, the paper also addresses common concerns about LLMs and identifies future research opportunities in urban studies. This comprehensive analysis aims to provide valuable insights for researchers exploring the application of LLMs in urban studies and for those who have yet to begin utilizing them in their research.