This deliverable presents methods employed in LinkedTV to create, update and formalise a semantic user model that will be used for concept and content filtering. It focuses on the ex-traction of lightweight and dense implicit knowledge about user preferences. This process includes the semantic interpretation of information that stem from the user’s interaction with the content, together with the estimation of the impact that pre-ferred concepts have for each specific interaction based on the type of transaction and the user’s physical reaction to the con-tent. User preferences are then updated based on their age, frequency of appearance and utility, while persistent associa-tions between preferences are learnt. This information evolves to a semantic user model that is made available for predictive inference about relevant concepts and content.