EN (which powers the Rdio create station feature that rolled out recently) has similarity data and popularity data for songs & artists natively.

Popularity of a single song: http://developer.echonest.com/api/v4/song/search?api_key=FIL...

List of songs by an artist ordered by popularity

http://developer.echonest.com/api/v4/song/search?api_key=FIL...

Since we have Rdio in our Rosetta ID space, you can natively use Rdio IDs and get them back in your calls:

http://developer.echonest.com/api/v4/song/search?api_key=FIL...

#awesomesauce

Music listening is completely subjective based on the smallest of changes within any given moment. Sure, the ideas of play strength/stars/constant radio/genre/BPM all helps in culling a mood, but it will never go all the way (without some sort of brain connection device).

I may create a playlist one day based upon a certain artist (with similar artists in the mix) however the next day I may want to listen to something a bit heavier, making those playlists irrelevant no matter how far I've drilled down my preference list.

While Play Strength is a good concept, it's still just a feature that can become just as underused or overused as a star count.

Instead of focusing on how 'perfectly-automatic' we can make a playlist based on features such as genre, BPM & play strength, we should be focusing on tools that help better understand our moods, the environment around us & the energy and vibes we intend to create from the music itself. The Effect > The Cause.

Selecting music to play at any given moment is entirely human. Play Strength was not intended to make music selection automatic.

I think humans can make better decisions with data. Play Strength attempts to expose insights already locked inside your music library.

For example, one of the big changes in driving some hybrid vehicles is a real-time monitor of how power is flowing in and out of the battery. I think just seeing that data changes your driving habits.

That's what I was going for with Play Strength. Thanks for sharing your thoughts.

> it's still just a feature that can become just as underused or overused as a star count.

I find star counts really useful. When you've got a huge amount of music it's hard to remember all the songs you like, and ratings (often in conjunction with other properties like genre) are useful way to filter the music.

Killer feature(idea): Work with a Kinect sensor on a feedback loop to update the playlist based on number of people, size of the room, acoustics, aggregate playlist taste of people who "check-in" to the room so their tastes gets included (you're a good host right?). I imagine it would bring a more social component to the playlist and put that social component to use.

You're blowing my mind here. Really good stuff.

What I think is really cool is that all this stuff is coming. There are a lot of stakeholders deeply invested in seeing these ideas come to fruition.

To the future! Cheers.

Very neat ideas here. I love the idea of helping Rdio choose what to play next by this combination of your recent plays combined popularity as defined by other users.

I also appreciate that many of the UI changes for these features are very subtle.

I just wonder how many users would actually find these features useful. I had to explain the concept of the "Collection" of a few people recently, which scared me since it's one of Rdio's core features!

Thanks for the kind thoughts!

I agree, the whole concept of a cloud-based music library is very new. Forget music, the concept of the cloud is new! I'm looking forward to a holiday filled with questions like, "where is it, who owns it, is it secure?"

Once people are comfortable with the paradigm shift, will really be cooking with gas.

I once had the idea of using markov chains to help build playlists. I find the order of songs can be very important, and that way I could build a more random playlist based liking song B after song A, and just keep going.

The problem was two fold, one, I really love whole albums when done right. So that most of the music I listen to is album based. The markov chain is pointless then. Two, I would need a really big data set to make it worthwhile. Rdio has that data set (or possibly even Echo Nest).

I love posts like this having spent a decent bit of time designing a music queuing UI myself. I think number 4 is really interesting if you could get the recommendations right. And you could do it in addition to play strength.

I like the idea of using stars to indicate favorite tracks. iTunes has a 5 star rating system that I use to keep my music organized and I wish Rdio had the same.

I need something more binary. Deciding between three and four stars seems totally arbitrary.

It's all about optimizing recommendations. If you just have a star or not starred. There's a chance the service will recommend a song you will likely hate (now you could be clever here and factor in skips). So now we have Hate, Love, Unrated. This works except for your OCD users will want to rate everything. And the more ratings you have, the better your recommendations will be.

What about Hate, Like, Love and unrated?

I always liked the idea of being able to specifically call out those few liked tracks that are a just a little more special :)

I only use one, three, and five stars, and it determines the "mood" of the song. One star songs are slow and chill, Three star songs are upbeat, and five star songs you'll want to rock out to.

Couldn't that be accomplished using BPM? Various tools will determine BPM for you.

Agree. Binary on/off with the stars is a lot easier to keep track of than remembering how you define each of the 1-5 star ratings.

All told, I think rating songs in your library is a really personal thing.

@JimEngland, you use stars purely as an organizational tool, which is slick. It's cool that the iTunes interface extends it self to different uses.

@tkahnoski and @zeedog, how would you elegantly expose controls for those different feelings?