摘要：Recommending to groups is even more complicated than recommending to individuals. Previous works has suggested that when generating recommendations to a group, it can achieve better result by learning information from other groups. Besides, recent research reports indicate that incorporating disagreement is critical to the effectiveness of group recommendation. Although the computation model build with Bayesian networks for group recommender system is very straightforward, the computation is rather complex (even though using approximate technology). In this paper, we will first present a Bayesian networks based evolution group recommendation model where groups can learn from each other. Then, we not only propose a new group recommendation computation framework, but also propose a new satisfaction measure model to refine the group recommendations. We evaluate the performance of our approach on the MovieLens dataset. Experiment results show that our Bayesian networks based evolution approach which is an ensemble of the above three sub-components outperforms the baseline one.
