Great to see a new type of recommenders (long overdue).

A related problem is of exploitation vs exploration (being recommended something one likes vs recommended new but with high uncertainty if it will be enjoyable or not)

some of my work on this related topic of active learning in recommender systems:

presentation: https://www2.slideshare.net/nrubens/active-learning-in-recom...

book chapter: http://machinelearning202.pbworks.com/f/Rubens-Active-Learni...

a more basic primer (aimed at startups) https://www2.slideshare.net/nrubens/1-of-40-recommender-syst...

am happy to answer any questions