Dr. Rani Nelken

Bio:​ Rani Nelken is Director of Research at Outbrain, where he leads a team focused on profiling users’ content consumption preferences by applying NLP and ML methods to users’ content interaction history. Before joining Outbrain, he was a Research Fellow in CS at Harvard ​University, and worked at several other companies including IBM Research and Mercury Interactiv​e. He received his PhD in CS from the Technion in 2001.

Dynamic online profiling of users’ content consumption preferences

Abstract​: Serving personalized content recommendations requires automatically creating user profiles reflecting users’ content consumption preferences and continually evolving them. Our profiles use abstract content-specific features of the consumed content, such as categories, topic models, and entities, which we automatically extract using NLP methods. Aggregating these features per user at scale raises interesting architectural and algorithmic challenges. In particular, it is impractical to store the entire dynamic history of a user’s interaction features, requiring us to use algorithms that selectively decay information in favor of a more compact representation. The talk describes some of our methodology for creating, updating, and evaluating these user profiles at large scale.