Matching Node Embeddings Using Valid Assignment Kernels.

Graph kernels have proven to be a promising approach for tackling the graph similarity and learning tasks at the same time. Most graph kernels are instances of the R-convolution framework. These kernels decompose graphs into their substructures and sum over all pairs of these substructures. A more promising family of kernels are the assignment kernels, which compute a matching between substructures of two objects such that the total similarity between the matched substructures is maximum. In this paper, we present a kernel which compares graphs by computing an assignment of their node embeddings. After embedding the vertices of all graphs in a vector space, we construct a hierarchy of the vertices using a clustering algorithm. Based on this hierarchy, we define a kernel that computes an optimal assignment of the vertices of two graphs. The proposed kernel is evaluated on several graph classification datasets. Our results indicate that the proposed approach is competitive with traditional and state-of-the-art techniques.