Abstract

The comparison of objects from a semantic perspective is a key point to improve the performance of current data mining tools and knowledge-based systems. Several semantic similarity measures defined in the field of Computational Linguistics are now employed in knowledge-based processes. Those measures are able to compare pairs of terms by leveraging the knowledge of an external source, like an ontology or a corpus. However, in some cases, the objects to be compared are not associated to a single term but they are described with a list of terms. In this paper

This paper has been selected by the CCIA'2013 scientific committee for an extended version in a special issue on AICommunications journal.

we propose a new similarity measure for objects described with multi-valued categorical attributes, which combines a uni-valued semantic distance with an OWA aggregation operator. The appropriateness and usefulness of this distance has been tested in the task of unsupervised clustering of objects within a personalised intelligent recommender system, obtaining promising results.