Ditto.Town is a new music recommendation service. It works like this: You enter your current favorite track and Ditto.Town will suggest a piece of music that is as similar as possible to the track you entered. You can rate Ditto.Town’s suggestion on the degree of similarity to the original track or even suggest a new track. This way, the similarity of the music tracks to each other becomes closer and the results of Ditto.Town will get better and better.
I'm Rettward von Doernberg and I develop Ditto.Town because I'd like to have better music suggestions myself.
Maybe you are like me in that there is always one piece of music that you currently just can’t get enough of. Then comes the point where you have simply listened to the track too many times. You want to find the next great track but how do you do this?
Wouldn’t it be great if you could just enter your current favorite track somewhere and suggestions for very similar pieces of music would appear, so that you could have a new favorite track again right away?
I had a closer look at the currently available recommendation services. Very rarely did they manage to suggest music tracks that I found to be truly similar. Recommendations from last.fm were probably the closest.
It seems the problem is that recommendations are automatically generated based on user behavior. Users, however, cannot interact with the system themselves, nor can they suggest anything new. This finally led me to develop a recommendation service myself.
Once you have entered your favorite track, Ditto.Town will first make suggestions of music tracks that have already been highly rated by previous users. If there are none or only a few rated suggestions are found, Ditto.Town pulls suggestions from another recommendation service (last.fm). These suggestions can also be rated and will be available for future requests. Since they have been rated for similarity, they are better than the original recommendations from the other recommendation service.
Ditto.Town uses a computer technology derived from the brain: a so-called “neural network.” Like many complicated-sounding things, it’s actually quite simple: when users rate the similarity between two pieces of music, they make a connection between those two tracks and also determine how close the connection is. Over time, after many pieces of music are eventually connected, a “neural” network is formed. This can then be used to find truly similar tracks.
It is important that as many connections as possible are created between the music tracks/nodes/neurons, and that these are evaluated as frequently as possible. Ditto.Town thus gets progressively smarter – it “learns” and becomes “more intelligent” the more and the longer the service is used. Much like a person (hopefully) grows progressively more intelligent through experience.
Ditto.Town is also completely objective. Market dominance of large music companies does not factor in Ditto.Town’s recommendations. It would be quite possible, for example, that the song of an unknown band would be suggested over a current hit of a superstar. Simply because users found these two songs to be very similar. Theoretically, this could lead to unknown artists expanding their audience quite quickly.
And Ditto.Town is immune to manipulation. If someone were to deliberately suggest a false similarity to another popular piece of music because, for example, they wanted to plug their own unknown track, this would simply be corrected by the next users. This would happen so quickly that hardly any users would be aware of the unknown piece of music. So it isn’t worth the trouble to enter false information. The same characteristic can also be found in blockchain technology.
And finally: if you’d like to contribute to the further development of Ditto.Town and help with server costs, you are welcome to donate something here (thank you!).