Netflix Home Feed Personalization: A UX and ML Teardown

AI · 7 min read

Netflix layers human editorial decisions on top of algorithmic ranking to produce a home feed that balances novelty with known preferences. The system uses contextual bandits to select which rows appear and reinforcement signals (start, finish, binge) to refine both content ranking and artwork selection.

A key lever is artwork personalization—different thumbnails for the same title shown to different users based on inferred visual preferences. This micro-optimization drives discoverability without changing underlying content catalogs. The UX uses modal previews and immediate playback to reduce decision friction and to gather rapid preference signals.

There are trade-offs: over-personalization risks creating filter bubbles, while too much editorial curation can suppress serendipity. The teardown recommends transparency affordances like 'Because you watched X' and transient surfacing of surface-level genres to help users escape algorithmic loops.