Also good point about collecting feedback about how users react to the way the recommendations was made. You could have different algorithms generating recommendations for different users depending on this e.g. collaborative filtering for some users, neural network for others, ensemble for the rest etc.