The goal of this project is to develop a recommender system that derives song recommendations from an implicit music dataset provided by the streaming service Spotify. We implemented current baseline systems and then two advancements over the baselines: Feature Enhanced Matrix Factorization and Non-Linear Matrix Factorization. To compare these systems, we took the predicted songs for a given playlist and calculated the performance score based on the accuracy of those results. We then compared the results from these NDCG scores to determine which system performed the best for the given Spotify dataset. Based off of the results, we were able to draw conclusions regarding the design process for an effective recommender system for music data.
Worcester Polytechnic Institute
Major Qualifying Project
All authors have granted to WPI a nonexclusive royalty-free license to distribute copies of the work, subject to other agreements. Copyright is held by the author or authors, with all rights reserved, unless otherwise noted.