Tune an e-commerce recommender across popularity, item-based collaborative filtering and a hybrid. Balance relevance against catalogue coverage and novelty, handle cold-start users, and avoid the filter bubble. Press Run to play 20 rounds, or Step to go one round at a time.
An e-commerce recommender suggests items to users. It must balance relevance (precision) against catalogue coverage and novelty, while handling new users.
Collaborative filtering scores items from user-item patterns. Chasing precision alone collapses the catalogue into a filter bubble; novelty and exploration defend coverage; cold-start users need special handling.
Use a hybrid with a little exploration - it defends coverage and click-through while keeping relevance high.
