An Effective Partitioning Approach for Competitive Spatial-Temporal Searching (GIS Cup)

The competitive spatial-temporal searching (CSTS) problem finds many applications in the daily life, such as ridesharing, car parking, and EV charging, where the mobile agents search for stationary resources on a road network. A key issue of CSTS is how agents should choose their search path such that the search time is minimized. To solve CSTS, we are faced with two main challenges: (1) how to find an optimal space partitioning granularity such that the resource availability pattern of each region is well modeled; (2) how to design a route search algorithm to avoid the "herding" effect as the agents tend to adopt a common search strategy. In this paper, we propose a Spatial-Temporal Partitioning (STP) approach to cope with these two issues. First, we partition the search space into regions by considering both spatial and temporal information of the historical resource records, and compute a weight for each region. Second, we assign a shortest-travel-time path to each agent from its current location to a relatively popular region according to the current time. Extensive experiments are conducted on a real dataset, which show that STP outperforms the baseline algorithm about 24 to 29 seconds in terms of average search time, and about 38% to 54% in terms of average wait time. The source code is available at: https://github.com/Chriszblong/STP.