Abstract  Knowledge-based recommenders support users in  the identification of interesting items from large  and potentially complex assortments. In cases  where no recommendation could be found for a  given set of requirements, such systems propose  explanations that indicate minimal sets of faulty  requirements. Unfortunately, such explanations are  not personalized and do not include repair proposals which triggers a low degree of satisfaction and  frequent cancellations of recommendation sessions.  In this paper we present a personalized repair approach that integrates the calculation of explanations with collaborative problem solving techniques. In order to demonstrate the applicability of  our approach, we present the results of an empirical  study that show significant improvements in the  accuracy of predictions for interesting repairs