Home

Spotify-ed: Music Recommendation and Discovery in Spotify

Hi, my name is José Bateira and this is my MSc thesis, developed at INESC TEC.

Supervisors

• Fabien Gouyon [email protected]
• Matthew Davies [email protected]

Documents

• Thesis Report
• 2-page abstracts: en, pt
• Prototype

Abstract

Not so long ago, before the Internet boom, listening to or discovering new music was a challenge on its own. Now, with a few clicks users, can have in their hands such a vast music catalogue that a human mind cannot compute it.

There are dozens of online services that offer exactly that. Some focus on creation and/or generation of playlists, others try to expand their music catalogue even further, or focus on a more personalized music recommendation. And these systems present their results to the user with a list or a grid of music artists, for example.

However, lists or grids do not give the user enough information about the relation between the results. A possible solution to this problem is to represent the artist's similarities as a network of interconnected artists in a graph, where a node is a music artist, and each edge between them represents a strong connection. This is the concept that RAMA (Relational Artist MAps), a project developed at INESC TEC, uses. From a single search, RAMA draws a graph that helps the user to explore new music that can catch his/her interest in a much more natural way. When a user wants to listen to an artist's music, Youtube's stream is used. Although one can find a large catalogue of music in Youtube, this service is not music oriented and the sound quality is not adequate for a music streaming service. The use of Youtube stream for audio thus needs to be replaced, and Spotify can provide a quality stream and an accurate music catalogue.

To address this question, this thesis proposes a Spotify App for RAMA. Will a Spotify user experience a more pleasant and natural way of music discovery from this graphical representation of artist relations within Spotify, than its standard discovery mode (with grids)? This is the main question that this dissertation seeks to answer.