264 Measuring Graph Similarity Using Node Indexing and Message Passing Available to Purchase

In this paper, we present a generative model to measure graph similarity. The parameters of that random process generating the observed graph from a template are determined by the node indices, reflecting the structure compatibility between nodes. We propose to use the steady state of the dynamic system represented by a directed graph to index the nodes. A message passing algorithm using the loopy belief propagation is adopted to approximate the maximum a posteriori assignment (MAP) for the nodes. The resulted believes of self matching form a similarity measure of the nodes in a graph. Examples and experiments demonstrated that the proposed method worked well for large graphs.