Text this: Enhancing Robustness in Feature Importance Methods with NAFIC and CESHAP for Improved Interpretability