In this talk I will review some of our recent work on
collaborative filtering.
I will first briefly discuss probabilistic models based on matrix
factorization to model changing user preferences and
item popularity over time. I will also dive into some of our very
recent work on inferring user exposure (to items) from click data
including how we model this problem using the tools of causal
inference. I will conclude by discussing some future challenges
including opportunities and
challenges of using these models in online environments with real-time
constraints.

For all projects I will report empirical and exploratory
results using click data from the arXiv.org paper repository.