Many real-world graphs are temporal, e.g., in a socialnetwork persons only interact at specific points in time. Thistemporality directs possibledissemination processeson thegraph, such as the spread of rumors, fake news, or diseases.However, the current state-of-the-art methods for supervisedgraph classification are designed mainly for static graphs andmay not be able to capture temporal information. Hence,they are not powerful enough to distinguish between graphsmodeling different dissemination processes. To address this,we introduce a framework to lift standard graph kernels tothe temporal domain. We explore three different approachesand investigate the trade-offs between loss of temporalinformation and efficiency. Moreover, to handle large-scale graphs, we propose stochastic variants of our kernelswith provable approximation guarantees. We evaluate ourmethods on various real-world social networks. Our methodsbeat static kernels by a large margin in terms of accuracywhile still being scalable to large graphs and data sets. Thisconfirms that taking temporal information into account iscrucial for the successful classification of temporal graphsunder consideration of dissemination processes.Keywords—Temporal graphs, Classification, Kernel

Top