Title :
Complex Recommendations

Organizers :

  • Oren Sar Shalom
  • Ido Guy
  • Flavian Vasile
  • Markus Schedl
  • Haggai Roitman

Abstract :

Recommender systems have become one of the key vehicles of user traffic both on the web and on mobile applications. In recent years, these systems have evolved to include many complexities that touch all their components, moving the resulting solution far beyond the traditional recommendation setup. We believe that the set of discrepancies between real-world applications and the idealized recommender systems model has not been properly studied and that a tutorial on the subject is sorely needed.

To this end, this tutorial spans four parts, which cover complexities in both input and output of real-life recommendation scenarios. In the first part, we present state-of-the-art methods that incorporate complex content (e.g., image or compound text). The second part breaks away from the top-k item recommendations and presents solutions for structured prediction. The third part discusses complex usage of users with items. Finally, the fourth part focuses on people-to-people recommendations, which introduces unique challenges stemming from the fact that both the recommended item and the recommendation target are actual individuals.