Abstract
Shortest distance queries not only find important applications in real-word systems, it is also the foundation for numerous other graph analysis problems. State-of-the-art techniques for such queries use 2-hop labeling, in particular, the Pruned Landmark Labeling (PLL) index is among the best performing for small world networks. However, PLL suffers from large label size and index computation time when the graph is large. In this paper, we propose two techniques to address the problem. The first technique is to limit the landmarks to vertices in a minimum vertex cover, and the second is to use a hybrid index by decomposing the graph into a core and a forest, and combining the PLL index for the core and a simpler distance labeling for trees. Extensive experiments with real-world graphs verified the effectiveness of these techniques.