Spotify Discover is great at giving you music suggestions similar to the music you already listen to but sometimes the best music discoveries come from experiencing a totally new genre, style, or artist thats beyond what you usually listen to. The goal of Cinuosity is to help you have those experiences more often. We achieve this by inciting randomness into the playlist generation process.
Check out the live site at cinuosity.com
Cinuosity is open source. Check it out on GitHub.
The user is offered a single control over the playlist generation algorithm through a "weirdness" level as shown below. A weirder level will result in more diverse and less popular results while a low weirdness will produce a playlist with more popular songs that the user is more likely to be familiar with. The model is completely probabilistic and will generate unique results on all weirdness levels.
After selecting your weirdness level and hitting discover playlist generation will begin. After a second or two you will be redirected to a new page displaying your new playlist that has been saved to your account. It will assign the playlist a random name so it can be uniquely identified. Hit the back button below to go back to the main page and generate more playlists.
I built Cinuosity to be a tool to help me find new and unique music on Spotify. You aren't supposed to enjoy all of the songs that that Cinuosity places in each playlist but the goal is to find at least one song that you enjoy and have that lead you on a new path. My reccomendatiton is to build a playlist and listen through. If you hear a song you don't like skip it and move on. Once you get to the end build a new playlist and see what you discover. So far I have discovered hundreds of new songs I enjoy that I would have never otherwise been able o find. You can take a listen to some of those songs below in a special playlist I built.
I gave a TEDx talk at Clemson University about the basics of modern music recommender systems to provide some insight into how the systems we use at Spotify, Pandora, and YouTube work. After this background, I introduce the issue of cyclical recommendations, how, once we use recommender systems that use our listening history as a predictor, we tend to keep getting recommended the same kinds of music leading to a very “safe” listening experience. This can be dangerous, since there are often many disparate styles we may enjoy, but never get recommended these. I talk about Cinuosity as one possible way to help combat this problem, as well as some general ideas to help develop the future systems. These systems will hopefully integrate user controllable parameters that let users shape the algorithms for their desires.