Large Language Model-Based Sentiment Analysis for Public Transit Satisfaction Using Trip Chain Data

This study proposes a methodological framework for analyzing public transit user satisfaction by processing unstructured, voice-derived trip chain data. Traditional survey methods often do not capture the contextual and real time aspects of user experience. To address this limitation, this study employs large language models and natural language processing techniques to analyze trip-related feedback collected from 5,000 public transit users. Sentiment scores were extracted from a station-labeled subset of the data set and applied to a broader corpus to assess satisfaction levels across subway stations in the Seoul Metropolitan Area. Using a combination of sentiment analysis and K-means clustering, this study categorized 102 subway stations into three groups: (1) High Satisfaction Station Group; (2) Low Satisfaction Station Group; and (3) Moderate Satisfaction Station Group. These clusters were the basis for grouping stations, and user feedback on transit experiences was summarized for each station within its respective cluster. This approach enables the identification of low-satisfaction transit stations requiring immediate attention, as well as moderate-satisfaction transit stations where service equity must be addressed. The proposed methodology offers a practical alternative to traditional survey-based assessments by leveraging unstructured public transit users’ data. With further refinement and institutional support, this approach can enhance data-driven, user-centered transit planning and contribute to broader improvements in urban mobility.