I think old school recommendations behave that way. You will need to introduce reinforcement learning into your reco in order to explore and exploit.