Fast Approximation of Shortest Path on Dynamic Information Networks

Computing shortest paths between a pair of nodes on a rapidly evolving network is a fundamental operation, and it is applied to wide range of networks. Classical exact methods for this problem can not extend to large-scale dynamic graphs. Meanwhile, existing approximate landmark-based methods can not be simply adapted for dynamic weighted graphs. In this paper, we consider four modifications on networks, including multi-edge insertions and deletions, weight increments and decrements. To address problems above, we present two improvements to existing methods: a novel method of indexing a weighted graph and high efficiency of approximating arbitrary pairs of nodes. Experimental results demonstrate that much faster and more precise estimation is achieved by our approaches.