Efficiently Discovering Regions of Interest with User-Defined Score Function

Abstract

Region of Interest (ROI) queries are of great importance in many location based services. However, the previous studies on ROI queries usually adopt either a simple spatial data model or a non-flexible enough query geometry, e.g., fixed-size rectangle. In this paper, to fix these drawbacks, we propose a new ROI search operator called Radius Bounded ROI (RBR) query. An RBR query retrieves a subset of spatial objects satisfying co-location constraints and maximizing a user-configurable score function at the same time. We formally prove that answering an RBR query is 3SUM-hard, which implies that it is unlikely to find a sub-quadratic solution. To answer the RBR queries efficiently, we propose three algorithms, PairEnum, BaseRotation and OptRotation based on novel geometric findings. In addition, the query processing technique we proposed can be easily extended to other related problems like top-k ROI search. To demonstrate both efficiency and effectiveness of our proposed algorithms, we conduct extensive experimental studies on both real-world datasets and synthetic benchmarks, and the results show that OptRotation, our most efficient algorithm, achieves more than $10^3\times$ efficiency improvement on both real and synthetic datasets compared with the baseline algorithm.