Formal Models for Learning of User Preferences , a Preliminary Report

We present a preliminary report about formal models for the task of learning user preferences. Learning user preferences by examples is described under content-based learning models, such as the Mistake-bound Model and the Probably Approximately Correct (PAC) model of Computational Learning Theory. Then we describe the learning of user preferences under a collaborative-based learning schema inspired by Collaborative Information Filtering and Re-commender Systems. Finally we formalize the problem as a joint learning problem under the Co-training model [Blum and Mitchell 1998], in order to combine content and social information for a more accurate model of human preferences. As result, different concepts and methods of learning user preferences can be expressed under a unified formal framework.