Temporal and Spatial Dynamics of Public Sentiment Toward Indoor Ventilation in New York City: Evidence From Social Sensing and Large Language Models

Abstract

This study examines public sentiment toward indoor ventilation in New York City using social sensing and large language models applied to geotagged Twitter data from 2013 to 2023. A longitudinal corpus of 1,500 tweets was compiled and analyzed to identify temporal and spatial patterns in ventilation-related discourse across New York City boroughs. After data cleaning and relevance screening, sentiment classification was conducted using RoBERTa to distinguish positive and negative expressions associated with indoor ventilation. The results show that discussion of indoor ventilation remained limited before 2019 but increased sharply during the COVID-19 period, with the highest volume observed in 2020 and sustained activity through 2022. Spatial analysis further indicates that Manhattan accounted for the largest concentration of ventilation-related tweets, followed by Brooklyn and Queens, suggesting uneven borough-level engagement with indoor air issues. Model confidence scores remained relatively stable across years, supporting the consistency of the classification results. Overall, the findings demonstrate how Twitter, as a form of social sensing, can help reveal changing public perceptions of indoor ventilation in urban environments, while large language models provide an effective approach for interpreting sentiment in geolocated public health discourse.