Article Preview
Top1 Introduction
Semantic similarity and semantic relatedness are two important concepts, which find their application in many areas, from Natural Language Processing (NLP) and Information Retrieval to Semantic Web. Semantic similarity usually accounts for common features in concept definitions, whereas semantic relatedness takes into account any functional or lexical relation between concepts as well (Hirst & St-Onge, 1998). Usually, semantic similarity considers only subsumption relation (mammal-cat), whereas semantic relatedness considers other types of relations, such as hyponymy/hypernymy, meronomy/holonymy, etc., as well as any other kind of functional relationship or frequent association (cat-milk). Hence, semantic similarity can be seen as a special case of semantic relatedness (Resnik, 1995) and is much more difficult to compute.
In this work we look at the similarity and relatedness of concepts in domain ontologies, where concepts are distinguished by their position in a taxonomy derived from the ontology and by the properties associated to them. Ontologies (Gruber, 1993; Guarino & Poli, 1995) are widely adopted to represent and model domain knowledge in web applications, and lot of attention is paid to the development of new ontology languages and associated reasoning mechanisms. Ontological representation of domain objects allows for representing both similarities and differences among domain objects as well as generic objects and very specific ones, allowing to define uniform classification criteria. Ontologies also include relations among objects and make it possible to derive implicit information from explicitly represented knowledge.
Drawing inspiration from Tversky’s work on Features of Similarity (Tversky, 1977), our working hypothesis on similarity and relatedness among ontological objects, where objects include concepts and their instances, is the following: two objects are similar if they both are defined having the same properties with the same values and two objects are related if they are defined having the same values for different properties. For example, in the event domain, Cooking_course and Dinner are similar concepts since both are defined having the same value Restaurant for the property has_venue_type. On the other hand, Football_game has the value Sport for the property has_main_argument, whereas Stadium has the same value Sport for a different property is_venue_for. The same happens for instances. For example, the two movies Hair and Amadeus are similar since they have the same value Forman for the property has_director, whereas the movies Desperately_Seeking_Susan and MTV_Video_Music_Awards are related since they have the same value Madonna for the properties has_main_actor and has_singer, respectively. Hence, the related objects are somehow connected, even when they are not similar.
We apply this semantic similarity and relatedness measure to the propagation of user interest values in ontology-based user models. A user model (Kobsa et al., 2001) is a knowledge structure which maintains users’ features (such as demographic features) and users’ attitudes (such as interest and knowledge) in the main domain objects. In an ontology-based user model, the user interest values are assigned to domain concepts and items as an overlay over a domain ontology (Brusilovsky, 2007). In these kinds of user models it is possible to exploit the ontological structure in order to propagate user values over ontology objects, starting from a small number of initial objects to other similar and related objects, and to incrementally update the user model (Cena et al., 2012; IJntema et al., 2010; Sieg et al., 2007).