How ZPlaylist Generator Curates Music Tailored to You
1) Inputs it uses
- Listening history (play counts, skips, favorites)
- Seed items (songs, artists, albums, or mood/keyword prompts)
- Context signals (time of day, activity tags like “workout” or “study,” and playlist length)
- Explicit preferences (genre, era, energy/tempo sliders)
2) Core personalization steps
- Profile embedding: converts your listening history and preferences into a compact user taste vector.
- Item representation: maps tracks and artists into the same embedding space using audio features (tempo, key, timbre), metadata (genre, release year), and popularity signals.
- Similarity & relevance ranking: finds candidate tracks nearest your taste vector and seed items, then ranks by relevance, novelty, and freshness.
- Diversity filtering: applies constraints to avoid repetition (artist/album caps, genre spread) and introduces serendipitous recommendations.
- Contextual tuning: adjusts selections for context (e.g., higher tempo for workouts, calmer tracks for evening).
- User feedback loop: updates the taste vector over time from saves, skips, and explicit edits to improve future generations.
3) Special features that improve fit
- Prompt-based generation: accepts natural-language prompts (“cozy indie for rainy mornings”) to steer mood and theme.
- “More like this” expansion: grows a playlist from one liked song or artist.
- Refresh schedules: auto-updates playlists (daily/weekly) to keep them fresh.
- Explainability: provides short reasons why specific tracks were chosen (e.g., “matches your love of 2010s R&B + mellow tempo”).
4) How it balances familiarity vs discovery
- Uses a tunable mix: a portion from known favorites (to ensure enjoyment) plus a fraction of novel/lesser-known tracks (to broaden taste). Parameters control how conservative or exploratory the playlist is.
5) What you can do to get better results
- Connect and allow history access for the streaming services you use.
- Give specific prompts (genre + mood + activity).
- Curate generated lists (remove/add tracks) so the algorithm learns your refinements.
If you want, I can produce three example prompts to generate distinct ZPlaylist outputs (e.g., workout, study, cozy evening).