The challenge focuses on recommending news articles in real-time. The challenge will traverse a two-stage process. In the first stage, we will provide a comprehensive data set. The data set will include both user and item features along with interactions in between them. Interactions can be characterized as either clicks (a user clicked on a recommended article) or impressions (a user reads an article). The data set will allow participants to tune their recommendation algorithms. In the second stage, participants will have the chance to directly interact with a real-time recommender system. The recommender system receives recommendation requests from various websites offering news articles. Requests are triggered by users visiting those websites. Participants will have access to a virtual machine where they can install their algorithm. The recommender system will forward the incoming requests to a random virtual machine. The random choice will be uniformly distributed over all participants. The virtual machine will have to provide recommendations. Alternatively, participants may setup their own server to respond to incoming requests. Note that there will be a fixed response time limitation. Hence, the participants will experience typical restrictions for real-world recommender systems. Such restrictions pose requirements regarding scalability as well as complexity for the recommendation algorithms. The recommender system will monitor the performance of all participants during the challenge duration. Participants may have the chance to continuously update their parameter settings in order to improve their performance levels. At the workshop, the performances will be presented and the best solution will be awarded.


  • July 1, 2013, data set release
  • August 15, 2013, start online competition


  • 1st place: iPad3
  • 2nd place: Nexus 7 inbeat
  • 3rd place: Raspberry Pi i2r
You may register for the online challenge with real-time user interactions via