Efficient K-NN for Playlist Continuation

Abstract

We present our solution for the RecSys Challenge 2018, which reached 9th place on the main track leaderboard of the competition. We developed a light-weight playlist-based nearest neighbor method to complete music playlists by using the playlist-track matrix along with track and playlist metadata. Our solution uses a number of domain specific heuristics for improving recommendation quality. One major advantage of our approach is its low computational resource use: our final solution can be computed on a traditional desktop computer within an hour.