Explainable Artificial Intelligence for Large Language Models: A Systematic Review and Evaluation Framework

This paper addresses the opacity of Large Language Models (LLMs) by presenting a critical review of Explainable Artificial Intelligence (XAI) techniques. Focusing on post-hoc methods, specifically feature attribution and counterfactual explanations, this work moves beyond experimental implementation to organize and analyze the existing literature. We examine current evaluation approaches, including fidelity and robustness metrics, to identify significant fragmentation within the field. Consequently, we propose a theoretical evaluation framework that structures the assessment of these methods. By highlighting key research gaps and future directions, this review provides a unified pathway for developing explainability standards that effectively balance technical rigor with human comprehensibility.