An Attributed Graph Embedding Method Using the Tree-Index Algorithm

Abstract

In this paper, we propose an embedding method for attributed graphs. For an attributed graph, we commence by using a tree-index method with the objective of strengthening the vertex labels. For each iteration of the tree-index method, we compute a probability distribution in terms of the frequency of the strengthened labels. With each probability distribution, we compute a Shannon entropy to measure the uncertainty of the strengthened labels. For an attributed graph, with the required Shannon entropies of different TI iterations to hand, we compute an entropy trace vector by measuring how the entropies vary with the increasing TI iterations (i.e., we embed the attributed graph into a vectorial space). We explore our method on several standard graph datasets abstracted from bioinformatics databases. The experimental results demonstrate the effectiveness and efficiency of our method. Our method can easily outperform state of the art methods in terms of the classification accuracy.