What we have done?

• From the perspective of transport service provider, we exploit user movement data to:
• Monitor Network-wide Traffic via Vehicle Trajectory Clustering (PVLDB’20)
• Bring People back to Public Transport via Reoptimizing the Bus Route Network (TKDE’19)
• Ehance Passengers’ Satisfaction via Optimizing the Network-wide Public Transport Time Schedule (DASFAA’20 and more on the way)
• Dynamic Ridesharing in Peak Travel Periods to Boost User Satisfaction (TKDE’20)
• From the perspective of individual user, we exploit user movement data to:
• Estimate the Origin-Destination Travel Time for a trip (SIGMOD’20)
• Build an All-purpose POI and Trajectory Search Engine for Personalized Trip Planning (ICDE’17, SIGIR’18, WSDM’18)
• Clean and Integrate Massive Trajectory Data (SIGMOD’18)

What problems we are solving?

• Answering the Exemplar Trajectory Query (ICDE 2017).
• A Unified Processing Paradigm for Interactive Location-based Web Search (WSDM 2018).
• Distributed In-memory Trajectory Analytics (SIGMOD 2018).
• A Search Engine for Trajectory Data on Road Networks (SIGIR 2018).
• Effective Travel Time Estimation: When Historical Trajectories over Road Networks Matter (SIGMOD 2020).

Demonstrations
FASTS
FASTS is a satisfaction-boosting bus scheduling assistant framework, which assists users to find an optimal bus schedule. FASTS performs bus scheduling based on the constraints specified by the user in either a coarse-grained or a fine-grained manner, supports different explorations with a varying number of constraints, and provides analysis to quantify the performance of bus schedules and presents the results in a visually pleasing way. We demonstrate FASTS using real-world bus routes and bus touch-on/touch-off records in Singapore.
Trip Planning
TISP is a Trip planning system by an Integrated Search Paradigm. TISP helps users (even those without any prior knowledge of the target city) interactively discover a city and incrementally plan a unique trip. Planning a trip usually involves a series of search processes, where users may issue several queries of the same type (with different settings), or even different types of queries, until the desired points of interest (POIs) and trajectories are found. In particular, for POI search, it involves the keyword query, k-Nearest Neighbor (kNN) query, Top-k Spatial Keyword (TkSK) query, Aggregate Nearest Neighbour (ANN) query, and Aggregate Textual Nearest Neighbour (ATNN) query. For trajectory search, it involves the k-Best-Connected-Trajectory (kBCT) query and Top-k Spatial-Textual Trajectory (TkSTT) query.
TTorch
We design a search engine, Torch, for trajectory data and serve almost all metric and non-metric trajectory similarity computations. In particular, we first map the trajectories into the road network, then a trajectory can be represented by a set of road segment ids. Moreover, a road segment is also crossed by many trajectories. Hence, we can use the inverted index to organize all the trajectories crossing a same road. Compression techniques over trajectories and inverted lists will further reduce the space of dataset and index.