Efficient quality-aware spatiotemporal data analytics

Abstract

Driven by the extensive emergence of spatiotemporal data and the prevalence of GPS-enable devices, the Location-Based Services (LBSs) achieve significant development, those refer to the applications integrating geographic locations with the general notion of services. Collectively, the proliferation of geo-positioning technologies raise many challenges in terms of research and industrial concerns. In recent years, many LBSs, e.g. route planning, points-of-interest search, and taxi-hailing service, tend to pursue not only the accomplishment of the service but also the quality of service (QoS). In particular, the QoS refers to the measurement for the overall performance of a service system based on users' satisfaction. For instance, the routing service strives to make use of the historical trajectories to recommend personalised paths according to user preferences; and the taxi-hailing service aims to optimise the system throughput further to reduce the users' waiting time. In this thesis, we identify a set of challenging problems regarding quality-aware LBSs and provide efficient solutions for them under various settings. Below is a brief description of our contributions.In order to improve the service quality for routing systems, we intend to mine users' different travelling behaviours from historical trajectories. Specifically, trajectories between the same origin and destination (OD) offer valuable information for us to better understand and mine the diversity of travelling preferences. However, due to the data sparsity issue, there are always insufficient trajectories to carry out mining algorithms for discovering the intrinsic properties of OD mobility. Thus, we propose an efficient and robust trajectory augmentation approach to reconstruct sizeable qualified trajectories from existing data. To explain, we concatenate existing trajectories to reconstruct a sufficient number of trajectories to represent the ones going across the OD pair directly, with a transition graph to support efficient sub-trajectory concatenation. In addition, we propose a novel similarity measure to calculate the distance between two sets of trajectories in order to validate whether the reconstructed trajectory set can well represent the original traces.Given a set of candidate nodes, a set of objects, e.g. points-of-interest, on the road network, and a distance threshold, the k-radius coverage query aims to return k nodes from the candidate set that can cover the maximum number of objects, where a candidate covers an object if the road network distance between them is no larger than the given threshold. A typical location-based application for this query is the best service spots selection for a certain service, e.g. gas stations, to maximise the scope of services for their users, enhancing their quality of service. This problem is NP-Hard and by a standard greedy algorithm, we can achieve a (1-1/e)-approximation guarantee. In this thesis, we improve the efficiency of the solution by a slight sacrifice on the approximation, allowing a small value of user-defined error parameter epsilon, in which case we are required to provide a (1-1/e-epsilon)-approximate solution. In particular, we propose two solutions: a reverse sampling based solution and a sketch-based approximation solution to address the k-radius coverage problem, where the query costs in both solutions are superior to the competitor by orders of magnitude.KNN queries for moving objects on the road network find many applications in LBSs, e.g. ride-hailing services, where each taxi is regarded as an object. To achieve high service quality, we propose a framework on conflict-aware kNN queries that optimises the system throughput and reduces the users' waiting time, while guaranteeing query correctness of query results under parallel processing. In such applications, objects are constantly moving such that even for the same query point, the correct answers of a kNN query may change over time. In addition, given a kNN request by a user, the k taxis returned by the query may all be occupied by concurrent queries requested by other users. Such conflicts cut down the system throughput and rise the query response time, leading to low QoS as a result. Consequently, this thesis presents a grid-based labelling with a scheduling framework that provides both efficient query time and low update cost whilst providing correctness guarantees for kNN queries.