SHAP explanations transfer robustly across architectures while LLM rationales transfer poorly to PLM predictions¶

Claim · All claims
Evidence · 1 study · causal r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? Removing LLM-ranked sentences produces substantially smaller PLM prediction shifts than removing SHAP-ranked sentences, while SHAP-ranked removal causes large immediate shifts in LLM predictions. → Ivo Bueno 2026

Evidence¶

Ivo Bueno 2026¶

Ivo Bueno, Babette Bühler, Philipp Stark, Tim Fütterer, Ulrich Trautwein, Dorottya Demszky, Heather Hill, Enkelejda Kasneci. (2026). From Scoring to Explanations: Evaluating SHAP and LLM Rationales for Rubric-based Teaching Quality Assessment. arXiv. https://arxiv.org/abs/2606.05180 doi:10.48550/arxiv.2606.05180

q2 · i? · causal · r2

Cross-model evaluation applying explanations from one model family to perturb the other (BERT large, DeBERTaV3 large, ALBERT base; Qwen3 235B, Llama 3.1 8B, Mistral Small). LLM deletion trajectories on PLMs were often non-monotonic; SHAP removal showed a consistent first-step jump on LLMs.

"In contrast, SHAP explanations consis- tently induce larger, more stable, and more coher- ent prediction shifts both within PLMs and when transferred to LLMs, demonstrating superior faith- fulness for sentence-level interpretability."

Discussion¶

Related Claims¶

• SHAP and LLM explanation methods show consistently low alignment, with mean top-10 Jaccard similarity of 0.085 — related
• Deletion tests show SHAP sentence rankings are more faithful than LLM-generated rationales for PLM scorers — possibly the same claim (merge candidate)