ML Basics and Tuning for a Streaming Recommender

MediumClassical ML

Reported in interviews at Tubi

Problem statement

How would you improve recommendations on a free video streaming service? A network retrieves titles from learned user and title representations, then a tree model ranks them by expected watch time. New titles arrive regularly, and the ranker learns from what the live row already shows.

Explain the models, training, evaluation, tuning, cold start, feedback loops and experiments.

Show hint

For the embeddings, count how many gradient updates a rare title's row receives compared with a popular one, and ask which optimizer adapts to that.

Relevant roles: ML Engineer, Data Scientist and Applied Scientist.

Related free problems