Causal inference for recommendation systems involves using techniques to estimate the causal effect of recommending an item on user behavior, such as engagement or purchase. Standard techniques include:
Instrumental Variable (IV) estimation: uses an instrumental variable to identify the causal effect of recommendation on outcome.
Propensity Score Matching (PSM): matches users who received recommendations with those who didn't, based on propensity scores.
Causal Trees and Forests: uses decision trees and random forests to estimate causal effects.