Constructing user preference profiles is crucial for accurate predictions in recommendation systems. While traditional profiling is well understood, modern dynamic online recommendation services raise interesting challenges for user profiling. First, effectively representing user preferences in dynamic settings requires novel modeling techniques such as latent factorization. Second, for space and time efficiency, it might be infeasible to store all observed user profile signals, which requires resorting to streaming or sketching algorithms. Third, there are significant engineering architecture challenges to efficiently maintain user profiles for large-scale real-time recommendations. We welcome papers addressing both algorithmic and architectural facets of the problem.
14:00-14:10 Opening introduction
14:10-14:50 Invited Talk: Laurent Charlin (HEC Montréal) Modeling User Preferences
14:50-15:10 Takuya Kitazawa. Incremental Factorization Machines for Persistently Cold-starting Online Item Recommendation
15:10-15:30 Jorge Díez, David Martínez-Rego, Amparo Alonso-Betanzos, Oscar Luaces and Antonio Bahamonde. Metrical Representation of Readers and Articles in a Digital Newspaper
15:30-16:00 Coffee Break
16:00-16:40 Invited Talk: Jonathan Siddharth (Rover) Building User Interest Profiles in Real-time for a Content Recommender System
16:40-17:00 Megan Bingham-Walker and Richard Searle. Personalized User Profiling for Time-Dependent Recommendation of Structured Products
17:00-17:20 Ido Tamir, Roy Bass, Baruch Brutman, Yoram Dayagi, Guy Kobrinsky and Ronny Lempel. Powering Content Discovery through Scalable, Realtime Profiling of Users' Content Preferences
17:20-17:30 Wrap up