Relational Kernel Machines for Learning from Graph-structured RDF Data

Presented at: 8th Extended Semantic Web Conference (ESWC2011)

by Veli Bicer, Thanh D. Tran, Anna Gossen

The amount of semantic data available as RDF is large and increasing. Despites the increased awareness that exploiting this large amount of data requires not only logic-based reasoning but also statistics-based inference capabilities, only little work can be found for the latter. On semantic data, supervised approaches particularly kernel-based Support Vector Machines (SVM) are promising. However, obtaining the right features to be used in kernels is an open problem because the amount of features that can be extracted from the complex structure of semantic data might very large. Further, instead of a single one, combining several kernels that specialize on subsets of features can help to deal with eciency and data sparsity. This creates the additional problem of identifying and combining dierent subset of features and kernels, respectively. In this work, we solve these two problems by employing the strategy of dynamic propositionalization to compute a hypothesis, representing the relevant features for a set of examples. Then, an R-convolution kernel is obtained from a set of clause kernels derived from components of the hypothesis. The learning of the hypothesis and kernel(s) is performed in an interleaving fashion, using a coevolution-based genetic algorithm for the underlying problem of multi-objective optimization. Based on experiments on real-world datasets, we show that the resulting relational kernel machine improves the SVM baseline

Keywords: Genetic Optimization, ILP, Kernel Machines, RDF, Relational Learning

Resource URI on the dog food server:

Explore this resource elsewhere: