LLMs Outperform Traditional NLP in Social Media Analysis for Urban Planning

Our research explores the potential of large language model (LLM) in analyzing Twitter (Now known as X) data to understand public perceptions of New York City’s congestion pricing plan. The study demonstrated that the LLMs can assist planners in obtaining satisfactory results from social media analysis with less time and preparation compared to classic natural language processing models. The GPT-4 model is the best of the four LLMs tested. Twitter users living farther from downtown tend to oppose the plan as the toll will increase their commute costs. Women, minority groups, and individuals in low-income occupations are underrepresented on Twitter, necessitating that planners consider alternative methods to solicit feedback from these demographics.

The voices behind congestion pricing: unveiling the social backgrounds of supporters and opponents through a large language model