Probabilistic Top-k Dominating Queries in Distributed Uncertain Databases (Technical Report)

TL;DR

This paper addresses the challenge of efficiently processing probabilistic top-k dominating queries on large-scale uncertain data in distributed environments, proposing a MapReduce framework with novel pruning and indexing strategies.

Contribution

It introduces a new distributed MapReduce framework for probabilistic top-k dominating queries on uncertain data, including pruning strategies and cost-model-based index distribution.

Findings

The proposed approach is efficient on real and synthetic datasets.

It effectively filters false alarms in distributed uncertain databases.

Experimental results demonstrate high effectiveness and scalability.

Abstract

In many real-world applications such as business planning and sensor data monitoring, one important, yet challenging, the task is to rank objects(e.g., products, documents, or spatial objects) based on their ranking scores and efficiently return those objects with the highest scores. In practice, due to the unreliability of data sources, many real-world objects often contain noises and are thus imprecise and uncertain. In this paper, we study the problem of probabilistic top-k dominating(PTD) query on such large-scale uncertain data in a distributed environment, which retrieves k uncertain objects from distributed uncertain databases(on multiple distributed servers), having the largest ranking scores with high confidences. In order to efficiently tackle the distributed PTD problem, we propose a MapReduce framework for processing distributed PTD queries over distributed uncertain databases. In this MapReduce framework, we design effective pruning strategies to filter out false alarms in the distributed setting, propose cost-model-based index distribution mechanisms over servers, and develop efficient distributed PTD query processing algorithms. Extensive experiments have demonstrated the efficiency and effectiveness of our proposed distributed PTD approach on both real and synthetic data sets through various experimental settings.