Fighting Spurious Correlations in Text Classification via a Causal Learning Perspective

We propose a method called the Causally Calibrated Robust Classifier (CCR) to mitigate spurious correlations in classification tasks.

Paper Code