How to avoid self-fulfilling prediction in recommendation systems?

bias
recommendation_engin
recommender

#1

I have two question on recommendation ML systems 1. If a ML system predicts that a user is likely to buy another item wouldn’t it mean that he is going to buy the recommended item anyway whether system recommends it or not? 2. If a ML system predicts that a user is likely to buy another item, then because it is shown so often compared to other things he could’ve bought the model will not have data to predict any new patterns. How to avoid this