One question that definitely has plagued me for a while is that of evaluation schemes of recommender systems... On multiple occasions I have designed recommender systems that give a fairly good user experience, but I just have not been able to quantify these results... any pointers on that will be really appreciated.