Graph kernels have become an established and widely-used technique for solvingclassification tasks on graphs. This survey gives a comprehensive overview oftechniques for kernel-based graph classification developed in the past 15 years. Wedescribe and categorize graph kernels based on properties inherent to their design,such as the nature of their extracted graph features, their method of computation andtheir applicability to problems in practice. In an extensive experimental evaluation, westudy the classification accuracy of a large suite of graph kernels on establishedbenchmarks as well as new datasets. We compare the performance of popular kernelswith several baseline methods and study the effect of applying a Gaussian RBF kernelto the metric induced by a graph kernel. In doing so, we find that simple baselinesbecome competitive after this transformation on some datasets. Moreover, we studythe extent to which existing graph kernels agree in their predictions (and predictionerrors) and obtain a data-driven categorization of kernels as result. Finally, based on ourexperimental results, we derive a practitioner’s guide to kernel-based graphclassification.

Top