Recommender systems often suffer from exposure bias, where we have customer feedback from only those items we actually recommend to the customers. Since only the shown items can collect positive feedback, we end up showing the same items again and again to the customers. This phenomemon is often called bias-amplification or a feedback loop.
This video talks about strategies to avoid feedback loops in recommender systems. Broadly three categories are discussed. (1) Designing algorithms to combat such bias (2) Feature engineering to avoid such bias (3) Strategies to re-order recommendations based on diversity, novelty and serendipity.