A family of novel graph kernels for structural pattern recognition

Abstract

Recently, an emerging trend of representing objects by graphs can be observed. As a matter of fact, graphs offer a versatile alternative to feature vectors in pattern recognition, machine learning and data mining. However, the space of graphs contains almost no mathematical structure, and consequently, there is a lack of suitable methods for graph classification. Graph kernels, a novel class of algorithms for pattern analysis, offer an elegant solution to this problem. Graph kernels aim at bridging the gap between statistical and symbolic object representations. In the present paper we propose a general approach to transforming graphs into n-dimensional real vector spaces by means of graph edit distance. As a matter of fact, this approach results in a novel family of graph kernels making a wide range of kernel machines applicable for graphs. With several experimental results we prove the robustness and flexibility of our new method and show that our approach outperforms a standard graph classification method on several graph data sets of diverse nature. © Springer-Verlag Berlin Heidelberg 2007.