CONFERENCE (INTERNATIONAL) Enabling Finer Grained Place Embeddings using Spatial Hierarchy from Human Mobility Trajectories
Toru Shimizu, Takahiro Yabe (Purdue University), Kota Tsubouchi
28th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems (ACM SIGSPATIAL 2020)
November 23, 2020
Forecasting anomalies in urban crowds is crucial for crowd control management. However, despite the success of recent pioneering work on urban anomaly forecast with mobility logs or transit search logs, accurate and long-term prediction of anomalous crowds is challenging because of limitations of existing methods. In this paper, we propose CityOutlook, an anomaly score matching-based method for accurate, long-term prediction of anomalous crowds. First, we formulate a regression model via data source association with mo- bility logs and transit search logs to leverage user’s schedules and the actual number of visitors. Second, we address the problem of dataset imbalance by oversampling strategy to ensure the model robustness. We evaluate CityOutlook using the datasets of large- scale mobility and transit search logs on daily urban dynamics and several anomalous crowds. The results show that CityOutlook can produce an accurate prediction of anomalous crowds 1 week in advance, and it outperforms the previous models both on anomaly score forecast and urban dynamics prediction.
Paper : Enabling Finer Grained Place Embeddings using Spatial Hierarchy from Human Mobility Trajectories (external link)