Mining and representing recommendations in actively evolving recommender systems

Abstract

Recommender systems provide an automatic means of filtering out interesting items, usually based on past similarity of user ratings. In previous work, we have suggested a model that allows users to actively build a recommender network. Users express trust, obtain transparency, and grow (anonymous) recommender connections. In this work, we propose mining such active systems to generate easily understandable representations of the recommender network. Users may review these representations to provide active feedback. This approach further enhances the quality of recommendations, especially as topics of interest change over time. Most notably, it extends the amount of control users have over the model that the recommender network builds of their interests.