Sunday, May 03, 2009

Latest Experiment - Recommender Bot

Image representing Blip as depicted in CrunchBaseImage via CrunchBase

For heavy Twitter users, we are all seeing the communication platform being used more and more heavily as a viral music/taste distribution tool. In my last post, I showed the traffic numbers of some of the more popular music services. One of the fastest growing at the moment is That information served as the inspiration of my latest mashup... I wanted to create an easy way to request recommendations to be pushed back to me based on a song that I "blipped".

So, a rainy weekend later I present to you the basic concept. Here's how it works.

  • just add the tag #recsplease to any "blip" (that gets published to Twitter)
  • the mashup monitors Twitter for blips that have the tag/recommendation request
  • from the URL I go and fetch/scrape the artist name for the song that was shared
  • this "seed" artist is then used to generate an array of similar artists
  • it constructs a tweet and replies to the requester with the list of recommendations

The next piece will be to link those recommendations off to a playlist, mix or radio station that includes those recommendations (in addition to trying to optimize the whole thing to be better, strong, faster). Maybe later this week. As always, let me know if you have any suggestions.

Reblog this post [with Zemanta]


Foafing your music said...

Hey Jason,

very nice (and useful) mashup. Well done!

BTW, how do you compute similar artists?

Cheers, Oscar

Stephan said...

cool mashup! music information retrieval meets social media ... same question as Oscar! :) details! Jaccard, Dice, Cosine ...

jherskowitz said...

Thanks guys, although I did not create my own recommendation engine... I'm just leveraging APIs of others. Right now I'm bouncing between Echonest and to fetch similar artists, although I may look at others as well.

Stephan said...

Hi Jason,

thanks! No shame about this :) "extreme mashup-ing" is more than ok!

Go ahead