Reproduce “Automatic Playlist Continuation through a Composition of Collaborative Filters” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Irene Teinemaa, Niek Tax, Carlos Bentes

Team Latte’s RecSys Challenge 2018 system combines track-, title-, album-, and artist-based collaborative filters to continue partially observed Spotify playlists. Category-specific weighting and an album-completion heuristic produced an official NDCG of 0.3612 and placed the system 12th among 112 teams.

Key insights

  • The best combination of recommendation signals depends on the available playlist context: title-only cases rely on title words, whereas seed-track cases are dominated by track co-occurrence.
  • Randomly sampled seed tracks represented playlist content better than merely using the first tracks; 100 random tracks achieved the strongest reported validation NDCG.
  • Album completion captures a structural playlist intent that standard collaborative scores may miss, while the popularity fallback guarantees the required list length but can introduce popularity bias.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.