Abstract
This study investigates the spatial and temporal variations in public sentiment towards electric vehicles (EVs) in the US, using Twitter data from 2019 to 2022. A hybrid data labeling approach is employed, which integrates large language models (LLMs) with human expert oversight to improve data annotation reliability. By addressing data imbalance with targeted resampling techniques and training multiple supervised learning models, this study provides an accurate analysis of public sentiment dynamics. Our findings reveal a significant spatial similarity between EV market share and the frequency of EV-related discussions on Twitter, suggesting that online discourse could serve as a potential indicator of EV market penetration. Furthermore, temporal sentiment analysis highlights an increase in negativity over time, particularly in regions with high EV adoption, revealing a contradictory trend between public satisfaction and the growing adoption of EVs. These insights offer valuable suggestions for policymakers and EV manufacturers, emphasizing the need to address consumer concerns, even in areas residents demonstrate higher willingness to adopt EVs, to foster more favorable public perceptions and support the broader adoption of EVs.