Maria Victoria Eyharabide - Ontology-based user profile learning

As the amount of data available over the Web is ever increasing, personalization techniques are needed so as to tailor the information delivered to users. Personalization requires extracting meaning out of all this raw data. Fortunately, a lot of effort is being put into extending the Web with semantic knowledge, such as ontologies, as an attempt to add this meaning. Even more, effective personalization involves two important challenges : accurately identifying the user’s relevant context and organizing the information in such a way that matches this particular context. In this talk, I will introduce a novel approach to building ontology segments to represent the user’s relevant context. In particular, I will show how to learn ontology segments using a combination of association rules, Bayesian networks and spreading activation. The context which is relevant to a particular user is learnt from his/her behavior’s observation and stored in a user profile. Then, I will detail a use case of this approach in which the user context is used to personalize the selection and display of information in an e-learning system. Furthermore, I will show how the proposed context learning strategy may be reused and combined to serve multiple purposes. Finally, I will describe the dynamical actualization of the user’s learnt context, and some future directions to extend this approach to social networks, based on collaborative personalization strategies.