Most existing music recommendation systems rely on listeners to provide seed tracks, and then utilize a variety of different approaches to recommend additional tracks in either a playlist-like listening session or as sequential track recommendations based on user feedback.

We built a playlist recommendation engine that takes a different approach, allowing listeners to generate a novel playlist based on a semantic string, such as the title of desired playlist, specific mood (happy, relaxed), atmosphere (tropical vibe), or function (party music, focus). Using a publicly available dataset of existing playlists (https://www.aicrowd.com/challenges/spotify-million-playlist-dataset-challenge), we combine a semantic similarity vector model with a matrix factorization model to allow users to quickly and easily generate playlists to fit any occasion.