Constructing CP-nets from Users Past Behaviors

While recommender systems over time have significantly improved the task of finding and providing the best service for users in various domains, there are still some limitations regarding the extraction of users’ preferences from their behaviors when they are dealing with a specific service provider. In this paper we propose a framework to automatically extract and learn users’ conditional and qualitative preferences by considering past behavior without asking any information from the users. To do that, we construct a CP-net modeling users’ preferences via a procedure that employs multiple Information Criterion score functions within an heuristic algorithm to learn a Bayesian network. The approach has been validated experimentally on a dataset of real users and the results are promising.