The Partition Bridge (PB) tree: Efficient nearest neighbor query processing on road networks

With the rapid development of location-based services, many types of applications in urban areas such as transportation planning, traffic management, and deployment of infrastructure depend on spatial objects that are distributed along the road network. As current place datasets include millions of spatial objects and human mobility datasets capture billions of locations, it is an important challenge to answer k-nearest neighbor queries efficiently. However, shortest-path distance calculations are the computational bottleneck for a k-nearest neighbor queries on road networks. Existing query algorithms partition the network to mitigate this bottleneck, but they do not take the data distribution into account, which leads to inefficiencies in dense areas and sparse areas. In this paper, an efficient and scalable indexing method called the Partition Bridge tree (PB-tree), is proposed based on hierarchical network partitions that consider network connectivity and data distribution. The structure of the PB-tree mainly includes distance matrices, union-bridges, bridges, and active network nodes component. Based on the PB-tree, a k nearest neighbor query processing algorithm is proposed by combining “bottom-up”, “top-down”, and adjacent extension methods. By using different road networks and datasets, the effectiveness and practicability of PB-tree are evaluated; the experimental results show that PB-tree outperforms the state-of-the-art methods and the classical approach.