ABSTRACT
Dynamic human activity intensity information is of great importance in many location-based applications. However, two limitations remain in the prediction of human activity intensity. First, it is hard to learn the spatial interaction patterns across scales for predicting human activities. Second, social interaction can help model the activity intensity variation but is rarely considered in the existing literature. To mitigate these limitations, we proposed a novel dynamic activity intensity prediction method with deep learning on graphs using the interactions in both physical and social spaces. In this method, the physical interactions and social interactions between spatial units were integrated into a fused graph convolutional network to model multi-type spatial interaction patterns. The future activity intensity variation was predicted by combining the spatial interaction pattern and the temporal pattern of activity intensity series. The method was verified with a country-scale anonymized mobile phone dataset. The results demonstrated that our proposed deep learning method with combining graph convolutional networks and recurrent neural networks outperformed other baseline approaches. This method enables dynamic human activity intensity prediction from a more spatially and socially integrated perspective, which helps improve the performance of modeling human dynamics.
Acknowledgments
Feng Lu acknowledges support from the National Key Research and Development Program (No. 2016YFB0502104). Mingxiao Li acknowledges support from the Guangdong Province Basic and Applied Basic Research Fund Project (No. 2020A1515111166) and a grant from State Key Laboratory of Resources and Environmental Information System. Hengcai Zhang acknowledges support from the National Natural Science Foundation of China (No. 41771436). Kang Liu acknowledges support from National Natural Science Foundation of China (No. 41901391) and Shenzhen Basic Research Program (No. JCYJ20190807163001783). Wei Tu acknowledges support from National Natural Science Foundation of China (No. 42071360). Song Gao acknowledges support from the National Science Foundation of United States (No. 1940091) and the Wisconsin Alumni Research Foundation. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the funders.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Data and codes availability statement
The data and codes that used in this research are available in figshare.com with the unique identifier at the link https://doi.org/10.6084/m9.figshare.11829306.v1. The data were packaged using the pickle package in Python. It included five parts: the human activity intensity array with a dimension of [8760,1614] and four kinds of interaction array with a dimension of [1614,1614], where 8760 indicates the length of human activity intensity series and 1614 indicates the number of cell phone towers. Since we signed a confidentiality agreement, we multiplied all the values by a random number from 0 to 1 to ensure that the original data were not revealed but could be used to demonstrate our method. The codes included our proposed AIP-IPS method and the foundation single-layer models of our proposed method. Our codes were based on Urban Computing Toolbox developed by Di Chai (https://uctb.github.io/UCTB/). The graph fusion technique was revised in the basic code to enable our model to consider the impact of integrated physical and social interactions on human activity intensity comprehensively.
Additional information
Funding
Notes on contributors
Mingxiao Li
Mingxiao Li is a Postdoctoral Researcher at Shenzhen University. He received his Ph.D degree from the Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences. His research interests focus on urban computing and spatiotemporal data mining.
Song Gao
Song Gao is an Assistant Professor in GIScience at the Department of Geography, University of Wisconsin-Madison. He holds a Ph.D. in Geography at the University of California Santa Barbara. His main research interests include Place-Based GIS, Geospatial Data Science, Human Mobility and Social Sensing.
Feng Lu
Feng Lu is a Professor with the Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences. His research interests cover trajectory data mining, computational transportation science and location-based services.
Kang Liu
Kang Liu is currently an Associate Professor in the Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences. She obtained her Ph.D. degree in Geographic Information Science (GIS) from the State Key Laboratory of Resource and Environmental Information Systems (LREIS), Institute of Geographic Sciences and Natural Resources Research (IGSNRR), Chinese Academy of Sciences (CAS) in 2018. Her research interests mainly focus on geographic big data and urban computing.
Hengcai Zhang
Hengcai Zhang is currently the Associate Professor at the Institute of Geographical Sciences and Natural Resources Research, Chinese Academy of Sciences. He received his Ph.D. degree from the Institute of Geographical Sciences and Natural Resources Research, Chinese Academy of Sciences in 2013. His research interest mainly focuses on moving objects database, urban computing and spatial-temporal data mining.
Wei Tu
Wei Tu is currently an Associate Professor in the department of urban spatial information engineering, Shenzhen University. His research interests include automatics recognition of human activity-mobility from multi-source urban data, trajectory modeling, and analysis and optimization.