A Human-Centered Interpretability Framework Based on Weight of Evidence