Concept Attribution on GoEmotions v1.0 (test)
42Average F1RISE
Evaluation Results
| Method | Links | |
|---|---|---|
| RISEBackbone=Qwen, Concept Category=Super Activators, Aggregation Protocol=LinSep2025.12 | 42 | |
| SHAPBackbone=Qwen, Concept Category=Super Activators, Aggregation Protocol=K-LinSep2025.12 | 40 | |
| SHAP IQBackbone=Qwen, Concept Category=Super Activators, Aggregation Protocol=LinSep2025.12 | 40 | |
| CALMBackbone=Qwen, Concept Category=Super Activators, Aggregation Protocol=LinSep2025.12 | 39 | |
| MFABABackbone=Qwen, Concept Category=Super Activators, Aggregation Protocol=LinSep2025.12 | 38 | |
| LIMEBackbone=Qwen, Concept Category=Super Activators, Aggregation Protocol=LinSep2025.12 | 37 | |
| GradCAMBackbone=Qwen, Concept Category=Super Activators, Aggregation Protocol=LinSep2025.12 | 36 | |
| LIMEBackbone=Llama, Concept Category=Super Activators, Aggregation Protocol=LinSep2025.12 | 34 | |
| MFABABackbone=Gemma, Concept Category=Super Activators, Aggregation Protocol=K-LinSep2025.12 | 34 | |
| FullGradBackbone=Qwen, Concept Category=Super Activators, Aggregation Protocol=LinSep2025.12 | 33 | |
| IntGradBackbone=Qwen, Concept Category=Super Activators, Aggregation Protocol=LinSep2025.12 | 32 | |
| Direct AlignmentBackbone=Qwen, Concept Category=Concept, Aggregation Protocol=LinSep2025.12 | 31 | |
| Direct AlignmentBackbone=Llama, Concept Category=Concept, Aggregation Protocol=LinSep2025.12 | 29 | |
| LIMEBackbone=Llama, Concept Category=Concept, Aggregation Protocol=LinSep2025.12 | 29 | |
| Direct AlignmentBackbone=Qwen, Concept Category=Super Activators, Aggregation Protocol=LinSep2025.12 | 28 | |
| Direct AlignmentBackbone=Qwen, Concept Category=Super Activators, Aggregation Protocol=K-Means2025.12 | 26 | |
| Direct AlignmentBackbone=Llama, Concept Category=Super Activators, Aggregation Protocol=LinSep2025.12 | 25 | |
| Direct AlignmentBackbone=Qwen, Concept Category=Concept, Aggregation Protocol=Avg2025.12 | 25 | |
| Direct AlignmentBackbone=Qwen, Concept Category=Super Activators, Aggregation Protocol=K-LinSep2025.12 | 25 | |
| Direct AlignmentBackbone=Gemma, Concept Category=Concept, Aggregation Protocol=LinSep2025.12 | 25 | |
| LIMEBackbone=Llama, Concept Category=Super Activators, Aggregation Protocol=Avg2025.12 | 25 | |
| Direct AlignmentBackbone=Qwen, Concept Category=Super Activators, Aggregation Protocol=Avg2025.12 | 23 | |
| Direct AlignmentBackbone=Qwen, Concept Category=Concept, Aggregation Protocol=K-Means2025.12 | 23 | |
| Direct AlignmentBackbone=Qwen, Concept Category=Concept, Aggregation Protocol=K-LinSep2025.12 | 23 | |
| Direct AlignmentBackbone=Gemma, Concept Category=Super Activators, Aggregation Protocol=LinSep2025.12 | 23 | |
| LIMEBackbone=Llama, Concept Category=Concept, Aggregation Protocol=Avg2025.12 | 20 | |
| Direct AlignmentBackbone=Llama, Concept Category=Super Activators, Aggregation Protocol=K-LinSep2025.12 | 19 | |
| Direct AlignmentBackbone=Gemma, Concept Category=Concept, Aggregation Protocol=Avg2025.12 | 19 | |
| Direct AlignmentBackbone=Llama, Concept Category=Concept, Aggregation Protocol=Avg2025.12 | 18 | |
| Direct AlignmentBackbone=Llama, Concept Category=Concept, Aggregation Protocol=K-Means2025.12 | 18 | |
| Direct AlignmentBackbone=Llama, Concept Category=Super Activators, Aggregation Protocol=K-Means2025.12 | 18 | |
| Direct AlignmentBackbone=Llama, Concept Category=Concept, Aggregation Protocol=K-LinSep2025.12 | 18 | |
| Direct AlignmentBackbone=Llama, Concept Category=Super Activators, Aggregation Protocol=Avg2025.12 | 16 | |
| Direct AlignmentBackbone=Gemma, Concept Category=Super Activators, Aggregation Protocol=Avg2025.12 | 16 | |
| Direct AlignmentBackbone=Gemma, Concept Category=Super Activators, Aggregation Protocol=K-LinSep2025.12 | 16 | |
| Direct AlignmentBackbone=Gemma, Concept Category=Concept, Aggregation Protocol=K-Means2025.12 | 15 | |
| Direct AlignmentBackbone=Gemma, Concept Category=Super Activators, Aggregation Protocol=K-Means2025.12 | 15 | |
| Direct AlignmentBackbone=Gemma, Concept Category=Concept, Aggregation Protocol=K-LinSep2025.12 | 14 |