Large Language Models As Standardization Tools: A Case Study Of Automatically Standardizing Transportation Data

The standardization of transportation data into uniform formats, such as the General Transit Feed Specification (GTFS), is crucial for ensuring interoperability, data quality, and effective decision making within the transport sector. This paper explores the innovative application of Large Language Models (LLMs) and Vision Language Models (VLMs), in automating the transformation of unstandardized data, including text and images, into standardized formats. Using the advanced reasoning capabilities of these models, we propose a methodology for extracting and standardizing transportation and mobility datasets. The focus of the work is in demonstrating the potential of LLMs in handling domain-specific standardization challenges. The paper details the development of a pipeline, from data collection and structured-data generation to the integration of human-in-the-loop to enhance model performance. On a set of 50 real-world timetable images, the best model, gemini-2.5-pro-preview-05-06, achieved mean±st.dev. GTFS-table scores of 4.78±0.58 (stops), 4.55±0.75 (trips) and 3.96±0.96 (stop_times) on a 1–5 scale.