Spotify generates natural-language taste profiles for millions to power LLM recommenders
_reachsumit · x · 2026-09-29
A Spotify team published Textual User Taste (TUT), a system generating natural-language taste profiles deployed to millions of users.
- Motivation: Behavioral embeddings remain strong for retrieval and ranking but are opaque and not natively suited to LLM workflows; foundation-model recommenders need user context that models can reason over and refine via natural-language interaction.
- Approach: Structured NL taste profiles are generated from listening behavior, interaction signals, content metadata, and optional user feedback, spanning a full production lifecycle including prompt development, compression, and user steering.
- Evaluation: Since no ground-truth taste profile exists, the team built a multi-faceted evaluation framework; profiles carry standalone predictive signal and improve recommendations when combined with behavioral embeddings.
More from Research
- Fully open-source VLA MolmoAct2 lands CoRL 2026 Spotlight, matching π0.5 performance — jmin__cho · 2026-09-30
- Image Lab Tops Robot Leaderboard: FLUX 3 Action Runs on a Desk-Size Arm — lmoroney · 2026-09-30
- MIT builds AI-powered Raman 'barcode' to noninvasively detect senescent 'zombie cells' — MacrinePhD · 2026-09-30
- Anil Seth's biological naturalism essay against AI consciousness draws mass commentary collection — anilkseth · 2026-09-30
- Causal Inference Assumptions: A Practical Checklist Before You Run the Model — Nasereliver · 2026-09-30
- Sparse-View Gaussian Splatting Demo Shows 3D Reconstruction From Few Images — enndeeee · 2026-09-30