Processing kNN Query with Pre-Computation in Time-Dependent Road Networks

KNN query in time-dependent road networks (TDk NN), which returns k nearest POIs with the fastest path from the query point at a specified departure time, are receiving increasing attention. Existing work for answering TDkNN expand vertices either in a Dijkstra-like way, or utilizing a heuristic function. They are not efficient especially in low POIs density cases since they cannot foresee where the next POI is located and have to expand the networks in every direction. In this paper, we pre-compute the INN results for every vertex in advance and store them as an index entry in order to guide the online expansion. An exact algorithm NNIDX-BF, which traverses vertices in a best-first style and could quickly locate where the next POI is located with the help of the pre-computed index, is proposed. Deriving from NNIDX-BF, three approximate algorithms are designed by combining with the heuristic function based expansion style and relaxing the stopping restrictions of NNIDX-BF. These algorithms could provide a good tradeoff between performance and accuracy for TDk NN query. Extensive experiments have been conducted to verify the performance of the proposed algorithms. The results show that both the expanded vertices and response time of NNIDX-BF are about 20% less than those of existing algorithm under various settings. The approximate algorithms reduce the expanded vertices by about 40% than existing algorithms with similar accuracy.