Concept Attribution on Sarcasm Dataset
0.74Average F1LIME
Evaluation Results
| Method | Links | |
|---|---|---|
| LIMEModel=Llama, Representation Target=Super Activators, Evaluation Protocol=K-LinSep2025.12 | 0.74 | |
| SHAP IQModel=Llama, Representation Target=Super Activators, Evaluation Protocol=K-LinSep2025.12 | 0.74 | |
| RISEModel=Qwen, Representation Target=Super Activators, Evaluation Protocol=LinSep2025.12 | 0.71 | |
| LIMEModel=Llama, Representation Target=Super Activators, Evaluation Protocol=LinSep2025.12 | 0.7 | |
| SHAPModel=Gemma, Representation Target=Super Activators, Evaluation Protocol=K-LinSep2025.12 | 0.67 | |
| CALMModel=Qwen, Representation Target=Super Activators, Evaluation Protocol=K-LinSep2025.12 | 0.67 | |
| GradCAMModel=Qwen, Representation Target=Super Activators, Evaluation Protocol=K-LinSep2025.12 | 0.65 | |
| FullGradModel=Llama, Representation Target=Super Activators, Evaluation Protocol=LinSep2025.12 | 0.65 | |
| MFABAModel=Gemma, Representation Target=Super Activators, Evaluation Protocol=K-LinSep2025.12 | 0.65 | |
| Direct AlignmentModel=Llama, Representation Target=Concept, Evaluation Protocol=LinSep2025.12 | 0.63 | |
| Direct AlignmentModel=Qwen, Representation Target=Concept, Evaluation Protocol=LinSep2025.12 | 0.58 | |
| IntGradModel=Gemma, Representation Target=Super Activators, Evaluation Protocol=LinSep2025.12 | 0.58 | |
| Direct AlignmentModel=Gemma, Representation Target=Concept, Evaluation Protocol=LinSep2025.12 | 0.57 | |
| LIMEModel=Llama, Representation Target=Concept, Evaluation Protocol=LinSep2025.12 | 0.52 | |
| LIMEModel=Llama, Representation Target=Super Activators, Evaluation Protocol=K-Means2025.12 | 0.5 | |
| LIMEModel=Llama, Representation Target=Concept, Evaluation Protocol=K-LinSep2025.12 | 0.5 | |
| LIMEModel=Llama, Representation Target=Super Activators, Evaluation Protocol=Avg2025.12 | 0.46 | |
| Direct AlignmentModel=Gemma, Representation Target=Concept, Evaluation Protocol=Avg2025.12 | 0.42 | |
| Direct AlignmentModel=Gemma, Representation Target=Super Activators, Evaluation Protocol=LinSep2025.12 | 0.4 | |
| Direct AlignmentModel=Llama, Representation Target=Concept, Evaluation Protocol=Avg2025.12 | 0.39 | |
| Direct AlignmentModel=Qwen, Representation Target=Concept, Evaluation Protocol=Avg2025.12 | 0.38 | |
| Direct AlignmentModel=Llama, Representation Target=Super Activators, Evaluation Protocol=LinSep2025.12 | 0.37 | |
| Direct AlignmentModel=Qwen, Representation Target=Super Activators, Evaluation Protocol=LinSep2025.12 | 0.37 | |
| LIMEModel=Llama, Representation Target=Concept, Evaluation Protocol=Avg2025.12 | 0.34 | |
| LIMEModel=Llama, Representation Target=Concept, Evaluation Protocol=K-Means2025.12 | 0.29 | |
| Direct AlignmentModel=Llama, Representation Target=Concept, Evaluation Protocol=K-Means2025.12 | 0.28 | |
| Direct AlignmentModel=Llama, Representation Target=Super Activators, Evaluation Protocol=K-Means2025.12 | 0.28 | |
| Direct AlignmentModel=Llama, Representation Target=Concept, Evaluation Protocol=K-LinSep2025.12 | 0.28 | |
| Direct AlignmentModel=Llama, Representation Target=Super Activators, Evaluation Protocol=K-LinSep2025.12 | 0.28 | |
| Direct AlignmentModel=Qwen, Representation Target=Super Activators, Evaluation Protocol=Avg2025.12 | 0.26 | |
| Direct AlignmentModel=Qwen, Representation Target=Concept, Evaluation Protocol=K-Means2025.12 | 0.26 | |
| Direct AlignmentModel=Llama, Representation Target=Super Activators, Evaluation Protocol=Avg2025.12 | 0.25 | |
| Direct AlignmentModel=Qwen, Representation Target=Super Activators, Evaluation Protocol=K-Means2025.12 | 0.25 | |
| Direct AlignmentModel=Gemma, Representation Target=Super Activators, Evaluation Protocol=Avg2025.12 | 0.25 | |
| Direct AlignmentModel=Qwen, Representation Target=Concept, Evaluation Protocol=K-LinSep2025.12 | 0.24 | |
| Direct AlignmentModel=Qwen, Representation Target=Super Activators, Evaluation Protocol=K-LinSep2025.12 | 0.24 | |
| Direct AlignmentModel=Gemma, Representation Target=Concept, Evaluation Protocol=K-Means2025.12 | 0.24 | |
| Direct AlignmentModel=Gemma, Representation Target=Concept, Evaluation Protocol=K-LinSep2025.12 | 0.24 | |
| Direct AlignmentModel=Gemma, Representation Target=Super Activators, Evaluation Protocol=K-Means2025.12 | 0.23 | |
| Direct AlignmentModel=Gemma, Representation Target=Super Activators, Evaluation Protocol=K-LinSep2025.12 | 0.23 |