No abstract

Abstract. The increasing availability of structured data in Resource Description Framework (RDF) format poses new challenges and opportunities for data mining. Existing approaches to mining RDF have only focused on one specific data representation, one specific machine learning algorithm or one specific task. Kernels, however, promise a more flexible approach by providing a powerful framework for decoupling the data representation from the learning task. This paper focuses on how the well established family of kernel-based machine learning algorithms can be readily applied to instances represented as RDF graphs. We first review the problems that arise when conventional graph kernels are used for RDF graphs. We then introduce two versatile families of graph kernels specifically suited for RDF, based on intersection graphs and intersection trees. The flexibility of the approach is demonstrated on two common relational learning tasks: entity classification and link prediction. The results show that our novel RDF graph kernels used with Support Vector Machines (SVMs) achieve competitive predictive performance when compared to specialized techniques for both tasks.

Abstract. Annotations of multimedia documents typically have been pursued in two different directions. Either previous approaches have focused on low level descriptors, such as dominant color, or they have focused on the content dimension and corresponding annotations, such as person or vehicle. In this paper, we present a software environment to bridge between the two directions. M-OntoMat-Annotizer allows for linking low level MPEG-7 visual descriptions to conventional Semantic Web ontologies and annotations. We use M-OntoMatAnnotizer in order to construct ontologies that include prototypical instances of high-level domain concepts together with a formal specification of corresponding visual descriptors. Thus, we formalize the interrelationship of high-and low-level multimedia concept descriptions allowing for new kinds of multimedia content analysis and reasoning.

Abstract. The amount of ontologies and meta data available on the Web is constantly growing. The successful application of machine learning techniques for learning of ontologies from textual data, i.e. mining for the Semantic Web, contributes to this trend. However, no principal approaches exist so far for mining from the Semantic Web. We investigate how machine learning algorithms can be made amenable for directly taking advantage of the rich knowledge expressed in ontologies and associated instance data. Kernel methods have been successfully employed in various learning tasks and provide a clean framework for interfacing between non-vectorial data and machine learning algorithms. In this spirit, we express the problem of mining instances in ontologies as the problem of defining valid corresponding kernels. We present a principled framework for designing such kernels by means of decomposing the kernel computation into specialized kernels for selected characteristics of an ontology which can be flexibly assembled and tuned. Initial experiments on real world Semantic Web data enjoy promising results and show the usefulness of our approach.

In this paper we propose a new approach to the design of semantic smoothing kernels for text classification. These kernels implicitly encode a superconcept expansion in a semantic network using well-known measures of term similarity. The experimental evaluation on two different datasets indicates that our approach consistently improves performance in situations of little training data and data sparseness.

No abstract

No abstract

Several Text Categorization applications require a representation beyond the standard bag-of-words paradigm. Kernelbased learning has approached this problem by (i) considering information about syntactic structure or by (ii) incorporating knowledge about the semantic similarity of term features. We propose a generalized framework consisting of a family of kernels that jointly incorporate syntactic and semantic similarity and demonstrate the power of this approach in a series of experiments.