Advanced query processing on spatial networks

Abstract

Recent advances in GPS and mobile communication technologies have allowed applications to emerge that can access and exploit location information about (moving) objects on road networks. Location-based services enable car drivers to search for facilities such as restaurants, shops, and car-parks close to their route. Logistic services monitor the status of delivery vehicles and ensure the timely delivery of goods. In this class of applications, both the accessibility and location of objects (e.g., vehicles and facilities) are constrained by the underlying network. The actual distance between two objects is defined by their shortest path distance on the network rather than their Euclidean distance. These network constraints significantly increase the complexity of retrieving spatial query results. Thus, query processing on spatial networks (i.e., road networks) has received considerable attention from database researchers in recent years. In this thesis, we identify three interesting problems and study their evaluation in the context of spatial networks: (i) aggregate nearest neighbor (ANN) query, (ii) reverse nearest neighbor (RNN) query, and (iii) clustering. Our findings for (i) and (ii) provide meaningful results for end-users, while our results for (iii) provide effective data exploration tools for data analysts. Aggregate nearest neighbor (ANN) queries are generalized from the nearest neighbor problem, allowing a group of mobile users to express individual preferences for reaching the best overall facility (e.g., a restaurant). Reverse nearest neighbor (RNN) queries are relevant to applications in decision support and resource allocation, enabling users to retrieve data objects locationally influenced by a query object. Clustering can be applied to discover dense collections of data objects, indicating regions of special interest. The process of computing results for these problems on spatial networks is complicated by the shortest path definition of the distance between two objects. Naive evaluation methods may lead to numerous expensive network distance computations, and may not scale well for large networks and large datasets. Our main research objective is the design of appropriate optimization techniques for the proposed problems, that incur low I/O cost of accessing the spatial network. We also investigate several variants of these problems in order to expand the application scope of our proposed techniques. Variants of ANN queries include aggregate center queries and weighted queries. RNN queries have bichromatic and continuous variants. Clustering is also applicable with several grouping criteria.