As stated in Wei et al. - Model-agnostic counterfactual reasoning for eliminating popularity bias in recommender system, Inverse Propensity Weighting (IPW) is a debias technique used in recommender systems that consists in decreasing the impact of popular items on the training, and increasing the impact of long-tail items with the purpose of reducing the Popularity Bias in recommender systems.
The problem with this method is the fact that it highly depends on the re-weighting strategy, which is notoriously hard to tune.
Because of this, other methods have been proposed during the years, by trying to counteract the popularity bias by analyzing the cause-effect of the items, by using Causal Graphs.
tags: recommender-systems