Constructing a Spatiotemporal Knowledge Graph for Urban Traffic from Trajectory Big Data

DescriptionThis dataset comprises data directly derived from, or further processed based on, the proposed urban traffic ST-KG in connection with the research presented in the paper. It is used to demonstrate the applicability of the proposed ST-KG across four key tasks: spatiotemporal analysis of congestion dynamics, traffic speed prediction, intelligent question answering on congestion, and tracing the causes of congestion. The data are organized according to these research tasks as follows:1. Urban Traffic ST-KG Construction & Congestion Level Assessment.zipContains the urban traffic ST-KG entities and relations (converted to CSV format) used for producing Figures 1 and 2 in the paper. 2. Spatiotemporal Analysis of Urban Traffic Congestion Dynamics.zipContains the processed results extracted from the urban traffic ST-KG and computed for exploring the spatiotemporal evolution of congestion. The dataset includes three folders—weekday, weekend, and holiday—each containing five subfolders (group 1–5) that store the average traffic speed of each grid cell within the corresponding group.3. Traffic Speed Prediction at the Regional Scale.zipContains the data used for traffic speed prediction, including the counts of seven types of Points of Interest (POIs) within each predicted grid cell, data from two precipitation stations, and the adjacency matrix of the predicted grid cells. The prediction target is the average traffic speed of the predicted grid cells, calculated at five-minute intervals. The file feature_matrix_X.csv stores the interpolated average traffic speed matrix.4. Intelligent Question Answering on Traffic Congestion.zipRecords the specific entities and relations extracted by the LLM-Agent from the urban traffic ST-KG in response to user queries.5. Tracing the Causes of Non-Recurrent Traffic Congestion.zipContains statistical analyses based on the results from Intelligent Question Answering on Traffic Congestion, further deriving traffic flow data. The files include:date_group_counts.xlsx – daily traffic flowdate_time_range_count.xlsx – traffic flow at five-minute intervalshoneycomb_time_date_count.xlsx – traffic flow of each grid cell at five-minute intervalshoneycomb_time_date_congestion_count.xlsx – traffic flow of each grid cell at five-minute intervals, categorized by congestion levelNaming NotesIn the paper, the "grid" is hexagonal in shape; therefore, it is referred to as "honeycomb" in the dataset.The "state" in the paper is derived from mapped trajectory points and is directly referred to as "trajectory_point" in the dataset.