A study of 5,000 Turkish tweets found that SHAP offered detailed feature attributions while local LLM explanations were ...