TrailMix: An Ensemble Recommender System for Playlist Curation and Continuation

Abstract

This paper describes TrailMix, an ensemble model designed to tackle the RecSys Challenge 2018 for automatic music playlist continuation. TrailMix combines three different models designed to exploit complementary aspects of playlist recommendation: (i) CC-Title, a cluster-based approach for playlist titles; (ii) DNCF, an extension of Neural Collaborative Filtering for taking advantage of the flat interaction among tracks; and (iii) C-Tree, a hierarchical approach akin to Phylogenetic trees for finding relationships between tracks.