Mining Constrained Regions of Interest: An Optimization Approach

摘要

The amount and diversity of mobile and IoT location and trajectory data are increasing rapidly. As a consequence, there is an emerging need for flexible and scalable tools for analyzing this data. In this work we focus on an important building block for analyzing location data, that is, the problem of partitioning a space into regions of interest (ROIs) that are densely visited. The extraction of ROIs is of great importance as it constitutes the first step of many types of data analysis on mobility data, such as the extraction of trajectory patterns expressed in terms of sequences of ROIs. However, in this paper we argue that unconstrained ROIs are not meaningful and useful in all applications. To address this weakness, we propose the problem of constraint-based ROI mining, and identify two types of constraints: intra- and inter-ROI constraints. Subsequently, we propose an integer linear programming formulation of the task of discovering a fixed number of constrained ROIs from a binary density matrix. We extend the approach to discover automatically the number of ROIs by relying on the Minimum Description Length Principle. Our experiments on real data show that the approach is both flexible, scalable and able to retrieve constrained ROIs of higher quality than those extracted with existing approaches, even when no constraints are imposed.