Projects

Equitable Bus Network Optimization

Public bus transport is a major backbone of many cities’ socioeconomic activities. As such, the topic of public bus network optimization has received substantial attention in Geographic Information System (GIS) research. Unfortunately, most of the current literature are focused on improving only the efficiency of the bus network, neglecting the important equity factors. Optimizing only the efficiency of a bus network may cause these limited public transportation resources to be shifted away from areas with disadvantaged demographics, compounding the equity problem. In this work, we make the first attempt to explore the intricacies of the equitable public bus network optimization problem by performing a case study of Singapore’s public bus network. We describe the challenges in designing an equitable public bus network, tackle the fundamental problem of formulating efficiency and equity metrics, perform exploratory experiments to assess each metric’s real-life impact, and analyze the challenges of the equitable bus network optimization task. For our experiments, we have curated and combined Singapore’s bus network data, road network data, census area boundaries data, and demographics data into a unified dataset which we released publicly. Our objective is not only to explore this important yet relatively unexplored problem, but also to inspire more discussion and research.

Spatiotemporal Similar Trajectory Search

Similar trajectory search is a cornerstone task in many spatial data analytic applications. Despite its importance, current literatures focus mostly on the spatial dimension of the data while paying little attention to the temporal aspect. Additionally, the few works that consider both aspects use point-to-point-based comparison methods that cannot adapt well to data and only use a single balancing factor to assign the importance between the spatial and temporal aspects. In this project, we utilize a deep representation learning approach that can adapt itself to different trajectories to find the proper balance between the spatial and temporal aspects, as well as learning a more in-depth representation of the trajectory points, both of which facilitate better performance. Experiments show that our model outperforms the current state-of-the-art deep-neural-network-based model as well as several spatio-temporal similar trajectory search baselines.