Spotify's Playlist Personalization: A Duel Between Taste and Serendipity

AI · 6 min read

Spotify blends user-level listening history, session context, and editorial curation to generate playlists that feel personal yet exploratory. Models include matrix factorization for long-term taste, sequence models for session intent, and contextual features like time of day or activity. The result is a layered system where different playlist products target different moments.

A key design decision is product separation: Discover Weekly acts as a weekly deep exploration tool, while Daily Mixes provide comfort and consistency. This separation reduces expectation misalignment; users know what type of surprise to expect. Spotify also uses human curation to inject the right kind of novelty and to bootstrap niche catalog gaps.

We recommend improvements around user controls for exploration intensity and better explanations for why tracks appear. From a product perspective, exposing sliders for familiarity vs novelty and a preview of why a track matched could give users more agency while preserving algorithmic serendipity.