"Hangzhou metro passenger flow dataset"

"There are approximately 70 million card-swiping transaction records collected from 81 metro stations across three lines, which are provided as the training data (Metro_train.zip). The dataset is organized in a day-by-day manner: each calendar day is stored as an individual CSV file, and filenames follow the pattern record_YYYY-MM-DD.csv. For example, all swiping records generated on January 1, 2019 (covering all three lines and all stations) are stored in record_2019-01-01.csv, and the same naming rule is used for the remaining days. Each record typically contains a timestamp and several identifiers (e.g., lineID, stationID, deviceID, userID), along with auxiliary fields such as status (e.g., entry\/exit type) and payType, enabling fine-grained aggregation of passenger flow by station and time interval.In addition to the transaction logs, a metro network map is provided in Metro_roadMap.csv, which specifies the topological connectivity of the metro system. This file encodes the adjacency relationships among stations (i.e., which stations are directly connected), and can be used to construct a station graph for downstream analysis\u2014such as defining neighborhood structures, modeling spatial dependencies, or building graph\/hypergraph regularization terms when recovering spatiotemporal passenger flow patterns."