Spatial-temporal data are data with both location and time dimensions, such as user trajectory, vehicular traffic, bike rental, etc. They not only reveal user mobility, preference and social activity, but also shed insights on the exchange of commodities or assets between regions. Spatial-temporal data mining hence plays a pivotal role in enabling many smart city applications, such as urban planning, traffic management, pandemic prevention and control, etc.

In the thesis, we first focus on the topic of co-location detection, which is to measure the spatial-temporal overlap between objects from their trajectories. It has many important smart city applications, such as contact tracing, companion detection, personalized marketing, etc. Given trajectories with explicit locations (e.g., geo...[ Read more ]

Spatial-temporal data are data with both location and time dimensions, such as user trajectory, vehicular traffic, bike rental, etc. They not only reveal user mobility, preference and social activity, but also shed insights on the exchange of commodities or assets between regions. Spatial-temporal data mining hence plays a pivotal role in enabling many smart city applications, such as urban planning, traffic management, pandemic prevention and control, etc.

In the thesis, we first focus on the topic of co-location detection, which is to measure the spatial-temporal overlap between objects from their trajectories. It has many important smart city applications, such as contact tracing, companion detection, personalized marketing, etc. Given trajectories with explicit locations (e.g., geo-location with latitude and longitude), we propose a novel and effective measure termed STS which considers location noise and heterogeneous data sampling rates in the data. Furthermore, for trajectories with strong location-privacy protection using WiFi Received Signal Strength Indicator (RSSI), we propose vContact, a novel, private, and effective IoT contact tracing solution given virus lifespan.

Then, we study the problem of flow forecasting using transition data between regions and aggregated flow data of any region, respectively. Flow forecasting is to predict the inflow (i.e., the number of trajectory destination located at a region per unit time) and outflow (i.e., the number of trajectory origin located at a region per unit time) for any region at the next time slot. It is crucial for many smart city applications, such as route planning, logistics and supply-chain management, public safety, etc. Leveraging transition data between regions, we propose a spatial-temporal graph neural network for docked bike prediction. Moreover, with only aggregated flow data of any region, we propose a spatial-temporal Transformer for flow forecasting.