Get on with it! - Recommender system industry challenges move towards real-world, online evaluation

AutorAndreas Lommatzsch, Jonas Seiler, and Daniel Kohlsdorf
QuellePresentation @ ECIR2016 Industry Day, Padua, Italy, March 23rd, 2016 
LinksDownload   |   BibTeX 

Recommender systems have enormous commercial importance in connecting people with content, information, and services. Historically, the recommender systems community has benefitted from industry academic collaborations and the ability of industry challenges to drive forward the state of the art. However, today’s challenges look very different from the NetFlix Prize in 2009. This talk features speakers representing two ongoing recommendation challenges that typify the direction in which such challenges are rapidly evolving. The first direction is the move to leverage information beyond the user-item matrix. Success requires algorithms capable of integrating multiple sources of information available in real-world scenarios. The second direction is the move from evaluation on offline data sets to evaluation with online systems. Success requires algorithms that are able to produce recommendations that satisfy users, but that are also able to satisfy technical constraints (i.e., response time, system availability) as well as business metrics (i.e., item coverage). The challenges covered in this talk are supported by the EC funded project CrowdRec, which studies stream recommendation and user engagement for next generation recommender systems.