Year: 2013
Type: dissertation
Abstract: In this thesis, we study query algorithms for location-based services (LBS) in road network distance. Nowadays, spatio-temporal database research and technology have promoted a diversion of LBS applications in the real world. For example, car navigation systems, pedestrian navigation systems, logistic systems, travel information systems are some well-known LBS applications. Additionally, various spatial query methods have recently been proposed for LBS applications. Some typical queries are nearest neighbor queries, range queries, spatial join queries, and trip planning queries (TPQ). For these queries, the target points of a query are points of interest (POIs) which are, for instance, restaurants, convenience stores, and gas stations. Moreover, because spatial data are multidimensional data, data access methods and algorithms for spatial queries become an important role. Therefore, the efficient indexing methods, R-tree indexing structure, for spatial query processing are basically applied to spatial databases. Spatial query methods were mainly based on Euclidean distance in the beginning, and then their focus shifted to the road-network distance. The difference between using Euclidean distance and road-network distance is in the calculation cost. For example, the Euclidean distance between two arbitrary points can be computed easily, however, in the road-network distance, it takes longer processing time. In the existing literature, to rapidly obtain the distance between two points, we can use some well-known shortest path finding algorithms such as Dijkstra's algorithm and A* algorithm for spatial queries in LBS application. These algorithms are well applicable for the shortest path queries between two points. However, when users want to get the shortest paths to multiple targets in parallel by using usual existing algorithms, the processing time becomes very long. Therefore, the simultaneous searching algorithms from a single source to multiple target points are practically necessary. In this thesis, we proposed a single source multi-targets A* (SSMTA*) algorithm to find the shortest paths to multi-target points simultaneously. This algorithm can control the unnecessary duplicated node expansions in spatial query than in the usual A* algorithm because the nodes are expanded at most once. We focused on the incremental Euclidean restriction (IER) strategy which rapidly generates a set of candidate POIs based on their Euclidean distance from a query point. And then, the Euclidean distance for each candidate is verified by computing their road network distance. At this road network distance computation, we applied the proposed SSMTA* algorithm. For trip planning queries such as optimal sequenced route (OSR) and trip planning query (TPQ), we proposed a fast incremental algorithm to find OSR candidates following IER framework. There is no previous study on the incremental search for OSR with IER framework. When IER framework is applied, OSR candidates in Euclidean distance are incrementally generates first. Let the shortest OSR given by searches in Euclidean space be Sr and its verified length in the road network be LN(Sr). The shortest OSR in the Euclidean space is not always the shortest OSR in the road network distance. Therefore, all OSRs whose lengths are less than LN(Sr) also have the potential to be the shortest route in the road network. All OSRs less than LN(Sr) must be searched in the Euclidean space, and then the results must be verified in the road network. Finally, the shortest OSR in the road network is returned as the result. These are the essential steps of an OSR query based on the IER framework. We also proposed an efficient verification method in the road network distance. Besides OSR with IER framework, we proposed algorithms based on the incremental network expansion (INE) for OSR queries. To evaluate the performance of the proposed SSMTA* algorithm, we applied it to several spatial queries; k nearest neighbor k-NN, aggregate nearest neighbor (ANN) queries. Besides, SSMTA* algorithm is applicable to other spatial queries in LBS applications. Firstly, the SSMTA* algorithm is applied to k-NN query in which k-NN candidates are incrementally generated in Euclidean distance. Then, these are verified in road network distance by SSMTA* algorithm. Through the performance evaluation, the proposed method compared with existing works; the INE using Dijkstra's algorithm, the pairwise A* algorithm and the lower bound constraint (LBC). The proposed method outperformed existing works in processing time and in expanded node number. The processing times of the pairwise A* algorithm and the LBC increase rapidly when the density of POIs is low or the number of k is large. On the pairwise A* algorithm, nodes can be expanded several times when k is large, which increases the hard-disk access times. On the LBC, although the number of node expansions remains low, the cost of PQ scanning increases in proportion to k and the number of node expansions. This performance deterioration is serious when the density of the POI is low. The expanded node number and processing time of Dijkstra's algorithm remains twice that of the proposed method. From the viewpoint of database systems, however, the number of disk accesses dominates the total calculation time. Accordingly, the proposed method is at least twice as efficient as the other methods. In applications to ANN queries, a set of k-number of ANN candidates are incrementally generated with Euclidean distances by using the minimum bounding method (MBM) method with the R-tree index. By applying the SSMTA* algorithm, the road-network distances from each ANN candidate point in the set to all query points are verified. We proposed three methods; ANNPQ, ANNQP and ANNQPLB for ANN queries and compared them with existing works. The intensive experiments show that SSMTA* algorithm outperformed the existing work in terms of processing time and expanded nodes number. For the simple trip planning (STP) query based on IER framework, the proposed the bi-directional distance constraint (BDDC) algorithm which based on the SSMTA* algorithm is compared with existing works. With various experimental results, we examined BDDC substantially outperformed previous methods in terms of both the expanded node number and the processing time. For OSR queries, we used two frameworks INE and IER. We proposed algorithms USVPG, BSVPG based on INE framework and evaluated. Comparing with the existing progressive neighbor exploration (PNE) method, the proposed methods is almost 100 times faster than PNE by thorough experiments. With IER framework, the proposed method EOSR query significantly outperformed PNE, particularly when POIs are densely distributed or the number of POI categories to be visited during the trip is large. We also proposed an algorithm to determine only one shortest route; however, the top k shortest routes are sometimes required to facilitate users' choices. The algorithm proposed in this study can be easily adopted for this requirement because the EOSR generates candidates incrementally and the algorithm for verifying the road network distance can be easily applied to k OSR queries.
Author: Htoo Htoo
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2266526954
