GenRec: Towards LLM-Native Recommendation at Netflix

Netflix's GenRec system aims for LLM-native recommendations, prioritizing long-term member satisfaction and business requirements like balancing diverse content (movies, series, games, live, podcasts) over immediate clicks. It uses catalog-aware scoring where each item has a learned embedding, and a scoring head combines this with a user representation to produce a score. These scores are then converted into a probability distribution and a ranking.