MSD Challenge Recommender Contest

April 26 started the contest of recommendation systems Million Song Dataset Challenge . Completion - in three months, on August 9. During the competition, you need to build a system that will be able to complete the missing 50% as accurately as possible with 100% of the history of listening to music for 1M users and 50% of the history for 100K users. At the same time, not only listening data is available, but also an extensive database of metadata and content data from The EchoNest, MusicXMatch and Last.fm. If you wish, you can use any other data (many other music services have an API through which you can extract valuable information).



The organizers are CAL UCSD , LabROSA CU , IMIRSEL and UIUC .



As such, the competition does not have a prize, but Zvooq decided to make it a little more interesting for Russian participants. The best team from Russia (regardless of the absolute location) will receive $ 5,000 and the opportunity to send one participant for free at ISMIR 2012.

')

Conditions for receiving this bonus - all team members must reside in the Russian Federation, the description of the approach used should be published (for example, at Habré or arxiv.org), the team must declare itself at challenge@zvooq.com.







PS As can be seen from the data provided and the objective function, which is used to evaluate decisions, a not quite recommendatory system is required - rather, you need to build a system that accurately guesses what the user has heard in the past.