Propagating User Interests In Ontology-Based User Models

With the advent of the Semantic Web, ontologies emerged as powerful tools for representing domain knowledge in many areas. In this thesis I will elaborate about their use for crafting user models in the context of recommender systems. A natural way to use ontology-based user models is to propagate the user interests in certain items to the related concepts, with goal of improving the quality of recommendations. This thesis will describe two main novel approaches for interest propagation and will also present an extensive evaluation of the different strategies and discuss the solutions which are best fit for different scenarios. In particular, I will introduce: 1) a novel approach to the anisotropic multidi- rectional propagation of user interests, particular effective on hierarchical ontolo- gies, 2) a novel approach to the property-based propagation of user interests using both semantic similarity and relatedness, and 3) the concept-biased cosine simi- larity, a similarity metric designed to exploit interest propagation for privileging certain concepts in an ontology and overcome data sparsity. These techniques will be evaluated according to the precision and diversity of their suggestions, showing that they outperform the state of the art solutions. The findings include that i) property based propagation outperforms the edge based one, ii) using also a relatedness metric can significantly increase the diversity of the results, and iii) these techniques can significantly improve the accuracy of a collaborative filtering system which suffer from data sparsity. Most of the material presented in this thesis derive from research papers pub- lished in international journals (Information Sciences) or international conference proceedings (UMAP, AI*IA, EKAW).