First Advisor

Douglas Hart


College of Computer and Information Sciences

Degree Name

MS Software Engineering

Document Type

Thesis - Open Access

Number of Pages

80 pages


Music streaming services use recommendation systems to improve the customer experience by generating favorable playlists and by fostering the discovery of new music. State of the art recommendation systems use both collaborative filtering and content-based recommendation methods. Collaborative filtering suffers from the cold start problem; it can only make recommendations for music for which it has enough user data, so content-based methods are preferred. Most current content-based recommendation systems use convolutional neural networks on the spectrograms of track audio. The architectures are commonly borrowed directly from the field of computer vision. It is shown in this study that musically-motivated convolutional neural network architectures outperform architectures that are highly-optimized for image-related tasks. A content-based recommendation model is built using musically-motivated deep learning architectures. The model is shown to be able to map an artist onto an artist embedding space where its nearest neighbors by cosine similarity are related artists and make good recommendations. It is also shown that metadata, such as lyrics, artist origin, and year, significantly improve these mappings when combined with raw audio data.

Date of Award

Summer 2019

Location (Creation)

Colorado (state); Denver (county); Denver (inhabited place)

Rights Statement

All content in this Collection is owned by and subject to the exclusive control of Regis University and the authors of the materials. It is available only for research purposes and may not be used in violation of copyright laws or for unlawful purposes. The materials may not be downloaded in whole or in part without permission of the copyright holder or as otherwise authorized in the “fair use” standards of the U.S. copyright laws and regulations.