Triangulation Versus Graph Partitioning for Tackling Large Real World Qualitative Spatial Networks

There has been interest in recent literature in tackling very large real world qualitative spatial networks, primarily because of the real datasets that have been, and are to be, offered by the Semantic Web community and scale up to millions of nodes. The proposed techniques for tackling such large networks employ the following two approaches for retaining the sparseness of their underlying graphs and reasoning with them: (i) graph triangulation and sparse matrix implementation, and (ii) graph partitioning and parallelization. Regarding the latter approach, an implementation has been offered recently, presented in [AAAI, 2014]. However, although the implementation looks promising and with space for improvement, an improper use of competing solvers in the evaluation process resulted in the wrong conclusion that it is able to provide fast consistency for very large qualitative spatial networks with respect to the state-of-the-art. In this paper, we review the two aforementioned approaches and provide new results that are different to the results presented in [AAAI, 2014] by properly re-evaluating them with the benchmark dataset of that paper. Thus, we establish a clear view on the state-of-the-art solutions for reasoning with large real world qualitative spatial networks efficiently, which is the main result of this paper.