Responsible Recommendation Systems through the Lens of Conformal Prediction and Causal Inference
Talk, USC ISI, Marina Del Rey, California
Recommendation systems play a crucial role in today’s massive information systems. Unfortunately, existing methods developed for maximizing utility of recommendation systems will lead to issues such as unfair item exposure allocation and popularity bias. In this talk, I will discuss two of our recent works. In the first work, we propose an efficient method for fairness utility trade-off in ranking, which transforms deterministic rankers to stochastic ones with finite-sample utility guarantees and improved fairness. In the second work, we formulate debiasing recommendation system as a multiple causal inference problem, and handle confounding bias by learning identifiable latent confounders.