Automated Playlist Continuation with Apache PredictionIO

Abstract

The Minrva project team, a software development research group based at the University of Illinois Library, developed a data-focused recommender system to participate in the creative track of the 2018 ACM RecSys Challenge, which focused on music recommendation. We describe here the large-scale data processing the Minrva team researched and developed for foundational reconciliation of the Million Playlist Dataset using external authority data on the web (e.g. VIAF, WikiData). The secondary focus of the research was evaluating and adapting the processing tools that support data reconciliation. This paper reports on the playlist enrichment process, indexing, and subsequent recommendation model developed for the music recommendation challenge.