DEEP DENSITY NETWORKS AND UNCERTAINTY IN RECOMMENDER SYSTEMS

Yoel Zeldes, Stavros Theodorakis, Efrat Solodnik, Aviv Rotman, Gil Chamiel, Dan Friedman

Abstract: Building robust online content recommendation systems requires learning com- plex interactions between user preferences and content features. The field has evolved rapidly in recent years from traditional multi-arm bandit and collabora- tive filtering techniques, with new methods integrating Deep Learning models that enable to capture non-linear feature interactions. Despite progress, the dynamic nature of online recommendations still poses great challenges, such as finding the delicate balance between exploration and exploitation. In this paper we provide a novel method, Deep Density Networks (DDN) which deconvolves measurement and data uncertainty and predicts probability densities of CTR, enabling us to perform more efficient exploration of the feature space. We show the usefulness of using DDN online in a real world content recommendation system that serves billions of recommendations per day, and present online and offline results to eval- uate the benefit of using DDN.

TL;DR: We have introduced Deep Density Network, a unified DNN model to estimate uncertainty for exploration/exploitation in recommendation systems.

Keywords: deep learning, recommendation system, uncertainty, context-based and collaborative filtering