How MixRadio’s “Natively Built Recommendation Engine” Works

MixRadio announced brand new & native recommendation engine, developed specifically to best deliver it’s one touch ‘Play Me’ personalized radio experience for a global audience. While it all sounds magical, when I read the news, the first question that I had—How is it exactly working and what it is.

Today MixRadio Dev Blog posted a lot of details about it. While I am not trying to decode here anything, but its is interesting to learn how it works, for consumers like you and me who started their day with MixRadio.

Lumia 830 MixRadio

How it used to work?

This might be the story of most of the Music Services when they started out. Most of the streaming services depend on third parties and brands to figure out what users are looking for. The same happened with Nokia Music and things were getting little slow as there is a limitation on what you can you can when you don’t have in-house data to analyze further.

The Evolution:

Things really changed when MixRadio as a brand came into picture. The Play Me button which delivers dynamic and personalised radio feature, and learns from what the user likes. This was still dependent on 3rd Party services and partners. So the team started working on their own, in-house recommendation engine which is live now.

So what is the new engine doing now

While there is no simple way to explain this, here is what I understand from the long and pretty detailed post from MixRadio Dev

MixRadio call this as Rex, which is basically a collection of back-end components, working with other services, generating dynamic and personalized playlists for our artist-seeded and Play Me radio use cases. You have two parts here — Analytics, which is offline and other is API and playlist algorithms which is online.

While the Analytics is busy crunching the data to create recommendations which comes from real time usage. It will help picking up good tracks for each artists for example. Next comes in the online part which uses the data to deliver on demand music experience. It is responsible for generating the dynamic sequences of tracks, refereed as play sessions.

So when you are playing a song, skipping and liking one, disliking another, the data is stored in the background and used in conjunction with above mentioned recommendations data and metadata as input to the playlist algorithms.

So now you see that instead of completely depending on a 3rd party services for everything, MixRadio Engine has data point for almost everything you do on the app. Right from downloading a mix to a skip. These form feedback, a strong basis for MixRadio to realize what you really like.

After all, if Play Me doesn’t start with a song I really like, it will make no sense for it trying to beome “your own personal radio station”.