ConcaveCubes: Supporting Cluster-based Geographical Visualization in Large Data Scale

Abstract

In this paper we study the problem of supporting effective and scalable visualization for the rapidly increasing volumes of urban data. From an extensive literature study, we find that the existing solutions suffer from at least one of the drawbacks below: (i) loss of interesting structures/outliers due to sampling; (ii) supporting heatmaps only, which provides limited information; and (iii) no notion of real-world geography semantics (e.g., country, state, city) is captured in the visualization result as well as the underlying index. Therefore, we propose ConcaveCubes, a cluster-based data cube to support interactive visualization of large-scale multidimensional urban data. Specifically, we devise an appropriate visualization abstraction and visualization design based on clusters. We propose a novel concave hull construction method to support boundary based cluster map visualization, where real-world geographical semantics are preserved without any information loss. Instead of calculating the clusters on demand, ConcaveCubes (re)utilizes existing calculation and visualization results to efficiently support different kinds of user interactions. We conduct extensive experiments using real-world datasets and show the efficiency and effectiveness of ConcaveCubes by comparing with the state-of-the-art cube-based solutions.