https://machinelearningmodels.org/improving-nlp-model-robustness/