RTSKG: Building a Rail Transit Station Knowledge Graph Dataset

Rail transit stations serve as critical hubs for urban mobility, yet traditional research often treats them as isolated entities or relies on small-scale, manually collected data. This approach fails to capture the complex, city-wide interactions—such as how a point of interest (POI) might influence ridership across multiple nearby stations. To address this, the authors propose the Rail Transit Station Knowledge Graph (RTSKG), a structured dataset designed to integrate heterogeneous urban data and model the spatial and semantic relationships necessary for city-level analysis.

The researchers constructed RTSKG for two major cities: New York City and Chicago. The construction process involved collecting four categories of urban data: rail transit systems (stations, lines, entrances/exits), administrative divisions (boroughs, functional areas), road networks (segments, categories), and POIs.

Central to this work is a custom-designed ontology that defines ten classes and nineteen relations. These relations are categorized into four types: geographic location, geographic adjacency, geographic intersection, and non-geographic relations. By using tools like GeoPandas, the authors extracted facts from these data sources and generated 'station areas'—isochrone regions representing areas reachable within 5 or 10 minutes of walking—to better model the local environment of each station.

The utility of RTSKG was evaluated through two specific downstream tasks: station-area store recommendation and knowledge-enhanced ridership prediction. By applying various Knowledge Graph Embedding (KGE) models, the researchers demonstrated that RTSKG provides a superior foundation for these tasks compared to existing urban knowledge graphs. The dataset is published as Linked Data, providing a standardized, scalable framework that can be extended to other cities by adapting the data collection and extraction scripts.