A Hierarchical Clustering Method for Semantic Knowledge Bases Nicola Fanizzi and Claudia d’Amato Dipartimento di Informatica, Universit`a degli Studi di Bari Campus Universitario, Via Orabona 4, 70125 Bari, Italy {fanizzi|claudia.damato}@di.uniba.it Abstract. This work presents a clustering method which can be ap- plied to relational knowledge bases. Namely, it can be used to discover interesting groupings of semantically annotated resources in a wide range of concept languages. The method exploits a novel dissimilarity measure that is based on the resource semantics w.r.t. a number of dimensions cor- responding to a committee of features, represented by a group of concept descriptions (discriminating features). The algorithm is an adaptation of the classic Bisecting k-Means to complex representations typical of the ontology in the Semantic Web. We discuss its complexity and the potential applications to a variety of important tasks. Key words: Description Logics, Hierarchical Clustering Algorithm, Medoids, Semantic Web 1 Introduction In the inherently distributed applications related to the Semantic Web (hence- forth SW) there is an extreme need of automatizing those activities which are more burdensome for the knowledge engineer, such as ontology construction, matching and evolution. An automatization of these activities may be achieved through the implementation of supervised or unsupervised inductive methods. In this work, we investigate on unsupervised learning for knowledge bases expressed in the standard ontological languages. In particular, we focus on conceptual clus- tering of semantically annotated resources. Essentially, clustering methods are based on the application of similarity (or density) measures, defined over a fixed set of attributes of the domain objects, with the goal of creating classes, namely homogeneous data subgroups. Classes of objects are taken as collections that exhibit low interclass similarity (density) and high intraclass similarity (density). Often these methods cannot take into account any form of background knowledge that could characterize object con- figurations by means of global concepts and semantic relationship. This hinders the interpretation of the outcomes of these methods which, on the contrary, is crucial in the SW perspective that foresees sharing and reusing the produced knowledge in order to enable semantic interoperability. Alternative approaches, particularly suitable for concept languages and terminological representations,