Songs can’t always be organized by BPM, key, or genre, and although Spotify’s algorithm works great for finding similar music, sometimes human involvement is what’s best for conveying the emotional value of songs. I began to think about how users could classify their music by activity, feeling, color, and so on, and use those categories as a means of communication.
Since I started using Spotify as my main streaming service, I've wanted a way to organize my music in a way that:
1. allows others to discover it, and
2. can communicate the emotions behind the music I love.
There are dozens of different ways that users can discover new music, but few that put power into the user's hands. Almost all means of discovery are through the platform itself, rather than user-to-user.
In order to build a better system of sharing and recommendation, I had to understand what was currently happening behind the scenes. Spotify uses three types of machine learning models to generate recommendations across the platform:
natural language processing models that analyze text-based metadata
collaborative filtering models that combine both your behavior and other users’ behavior
audio models that analyze raw audio tracks
natural language processing models that analyze text-based metadata
collaborative filtering models that combine both your behavior and other users’ behavior
audio models that analyze raw audio tracks
These models are used to create a Discover Weekly playlist every week as a means of introducing users to new music. As an example of how these models work, the Discover Weekly generally functions as follows:
In addition to Discover Weekly, Spotify has other avenues of discovery, some of which are shown below. These examples introduce users to new songs, playlists, and albums, and artists.
To gain a better understanding of how other streaming services handle music discovery, I ran a competitive analysis on SoundCloud and Apple Music. From looking at their interactions, information architecture, and features, I found that:
1. Although SoundCloud provides a tagging feature, it's only available to the users uploading the music, not to users listening to and saving the music.
2. SoundCloud has playlists for new and popular releases, but the suggestions aren't based on what the user has been listening to or interacting with.
3. Apple Music generates suggestions not only based on what you and users like you are listening to, but also what your friends are listening to.
Users can stumble upon other users playlists, but those playlists don’t have metadata associated to them to be able to seek them out by genre, mood, or activity.
There are many different ways of discovering new music, but all are sourced from the platform themselves—recommendations are through Spotify or Apple Music, rather than from user to user.
Users have to search for specific playlist names, which aren’t always directly related to the music they contain.
Users can stumble upon other users playlists, but those playlists don’t have metadata associated to them to be able to seek them out by genre, mood, or activity.
There are many different ways of discovering new music, but all are sourced from the platform themselves—recommendations are through Spotify or Apple Music, rather than from user to user.
Users have to search for specific playlist names, which aren’t always directly related to the music they contain.
Discovery.
The interface should allow users to unveil new songs and playlists according to genre, activity, and emotion. The interface should facilitate “falling into the rabbit hole” of new music.
Organization.
The design should enable users to organize their own libraries by genre, activity, and emotion.
Social connection.
The interface should facilitate users seamlessly sharing music with each other.
The design should create a space for users to feel emotionally connected through shared music.
A service that allows for more depth in organization and sharing may not be useful to the average Spotify user. This redesign would be targeted toward people passionate about music, particularly about organizing it in their own way and being able to share it with others—people who see their music library as an expression of themselves, and want to use it as a way to communicate in a way that’s often difficult with words.
I focused my user testing on the search results page, which I narrowed down to two options—searching with "#" and searching without it. I performed A/B testing to understand which information architecture users found more usable and easy to integrate into the existing UI.
After exploring and testing a few different options, the tagging function was deemed most usable when integrated into the ellipsis "more options" menu. This is an interaction users are already accustomed to, and, after A/B testing, proved to be easy to learn.
Songs tagged as "#tagName" will be sorted into two different spaces, the user-specific collection, and the platform-wide collection.
The user-specific collection enables users to organize their music in a new way, and can be shared in the same way as playlists. The platform-wide collection contains every song that Spotify users have tagged with "#tagName" assembled into a radio station.
.gif)
Scope and depth of discovery widens with platform-wide collaborative collections, and enables users to not only find new music, but also gain an understanding of how others interpret the tag. Tagged collections look and function similarly to playlists on Spotify's current interface.
When searching, ultimately, the users that I tested preferred including the "#" in the query to limit their search to tagged items, and wanted their own library prioritized over platform-wide collections.
It's hard to compete with a machine learning algorithm, even though music seems like something that's so deeply human. As I was working on this project, I found just shy of a dozen new albums through Spotify's avenues of discovery. That being said, the implementation of this design could open the door for more user involvement, and a wider scope of communication through music.
Diversity in user testing is crucial in making a usable product. I assumed everyone would like this feature, but in talking to friends and colleagues, it's most applicable to people who like to organize their library and find new music. Frankly, not everyone cares that much about music, which user interviews helped me to see.
It's hard to compete with a machine learning algorithm, even though music seems like something that's so deeply human. As I was working on this project, I found just shy of a dozen new albums through Spotify's avenues of discovery. That being said, the implementation of this design could open the door for more user involvement, and a wider scope of communication through music.
Diversity in user testing is crucial in making a usable product. I assumed everyone would like this feature, but in talking to friends and colleagues, it's most applicable to people who like to organize their library and find new music. Frankly, not everyone cares that much about music, which user interviews helped me to see.