Major project IT, Software & SaaS Case Studio

Playlist continuation and home-shelf ranking

Inspired by publicly reported work at Spotify

Suggest the next songs for a playlist and rank shelves on a home screen, evaluated like a real recommender.

The brief

Spotify has described ranking what appears on its home screen with machine learning, and it released a million real playlists for a research challenge. Build a playlist-continuation model on that data: given the first tracks of a playlist, suggest the rest. Then use the model to rank a small set of "shelves" for a listener and build a home-screen demo. Good looks like: candidate generation plus a re-ranker, R-precision and NDCG against a popularity baseline, cold-start handling for short playlists, and a working demo.

Read the case first

What Spotify has said in public about this kind of work, on its own pages:

Make it your own

Swap songs for something else people collect in lists: podcasts, YouTube study videos or recipes. Or keep music and add a rule of your own, such as giving new independent artists a fair share of slots, and measure what it costs.

AITeky has no link with Spotify, which has not reviewed or endorsed this brief. The brief only draws on what the company has published, uses open data rather than its data, and is a learning exercise: don't present your project as Spotify's work or system.

What you will build

Load the playlists and build track co-occurrence and embeddings
Generate candidates (co-occurrence, embeddings, popularity)
Train a learning-to-rank re-ranker
Evaluate with R-precision and NDCG; handle playlists with one or two tracks
Rank home-screen shelves for a listener
Build a demo and report diversity and popularity bias

Free, open data to build it with

It takes a minute with your email. AiTeky Pass (₹999/month or ₹9,999/year) is free for your first 30 days.