Subjectivity Classification on Subj (test)
97AccuracyFine-tuning (full)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Fine-tuning (full)Backbone=RoBERTa-large, Training Samples (K)=full, Evaluation Protocol=Full fine-tuning2021.08 | 97 | — | — | |
| Fine-tuned BERTSNN=false, Spike=false, PE=false, Param (M)=109.82024.05 | 95.9 | — | — | |
| Zhao et al. (2015)2017.09 | 95.5 | — | — | |
| DiSANevaluation=10-fold cross validation2017.09 | 94.2 | — | — | |
| SRU (4 layers)Size=303k, Optimizer=Adam, Word embeddings=fixed2017.09 | 93.8 | — | — | |
| RNN2018.05 | 93.7 | — | — | |
| Zhang and Wallace (2017)2017.09 | 93.7 | — | — | |
| SRU (8 layers)Size=502k, Optimizer=Adam, Word embeddings=fixed2017.09 | 93.7 | — | — | |
| SkipThoughtEvaluation Protocol=SentEval2018.08 | 93.7 | — | — | |
| HBMPEmbedding Dimension=1200D, Evaluation Protocol=SentEval2018.08 | 93.7 | — | — | |
| Skip-thoughtevaluation=10-fold cross validation2017.09 | 93.6 | — | — | |
| Wang and Manning (2013)2017.09 | 93.6 | — | — | |
| PLR-4Model=Llama3.1-8B-Instruct, k=322026.03 | 93.59 | — | — | |
| SRU (2 layers)Size=204k, Optimizer=Adam, Word embeddings=fixed2017.09 | 93.5 | — | — | |
| CNN2018.05 | 93.4 | — | — | |
| SRUevaluation=10-fold cross validation2017.09 | 93.4 | — | — | |
| Kim (2014)2017.09 | 93.4 | — | — | |
| QRNN (k=1)Size=165k, Optimizer=Adam, Word embeddings=fixed2017.09 | 93.4 | — | — | |
| QRNN (k=1) + highwaySize=204k, Optimizer=Adam, Word embeddings=fixed2017.09 | 93.4 | — | — | |
| KNN-Promptingprivacy budget (ε)=Non-private, number of hyperplanes=24, number of hash tables=42026.05 | 93.3 | — | — | |
| CAPOModel=Mistral-3.2-24B, Token Budget=7.5M2026.05 | 93.2 | — | — | |
| SWEM-concat2018.05 | 93 | — | — | |
| Wide CNNsevaluation=10-fold cross validation2017.09 | 92.9 | — | — | |
| CNNSize=360k, Optimizer=Adam, Word embeddings=fixed2017.09 | 92.7 | — | — | |
| HBMPEmbedding Dimension=600D, Evaluation Protocol=SentEval2018.08 | 92.7 | — | — | |
| LSTMSize=352k, Optimizer=Adam, Word embeddings=fixed2017.09 | 92.6 | — | — | |
| SWEM-aver2018.05 | 92.5 | — | — | |
| Spikformer w/ CPG-PESNN=true, Spike=true, PE=true, Param (M)=110.42024.05 | 92.5 | — | — | |
| InferSentEvaluation Protocol=SentEval2018.08 | 92.4 | — | — | |
| AdaSentevaluation=10-fold cross validation2017.09 | 92.2 | — | — | |
| Spikformer w/ Float-PESNN=true, Spike=false, PE=true, Param (M)=109.82024.05 | 92.2 | — | — | |
| MO-CAPOModel=Mistral-3.2-24B, Token Budget=7.5M2026.05 | 91.9 | — | — | |
| CAPOModel=Llama-3.3-70B2025.04 | 91.6 | — | — | |
| Spikformer w/o PESNN=true, Spike=false, PE=false, Param (M)=109.82024.05 | 91.6 | — | — | |
| LM-BFFShots=16/162021.07 | 91.4 | — | — | |
| Spikformer w/ Random-PESNN=true, Spike=true, PE=true, Param (M)=110.42024.05 | 91.4 | — | — | |
| cBoWevaluation=10-fold cross validation2017.09 | 91.3 | — | — | |
| SWEM-max2018.05 | 91.2 | — | — | |
| LM-BFFBackbone=RoBERTa-large, Training Samples (K)=16, Evaluation Protocol=Few-shot fine-tuning2021.08 | 91.2 | — | — | |
| CAPOModel=Qwen2.5-32B2025.04 | 91 | — | — | |
| CAPOModel=Qwen3-30B, Token Budget=7.5M2026.05 | 90.9 | — | — | |
| Fine-tuningBackbone=RoBERTa-large, Training Samples (K)=16, Evaluation Protocol=Few-shot fine-tuning2021.08 | 90.8 | — | — | |
| DARTBackbone=RoBERTa-large, Training Samples (K)=16, Evaluation Protocol=Few-shot fine-tuning2021.08 | 90.7 | — | — | |
| NSGA-II-POModel=Mistral-3.2-24B, Token Budget=7.5M2026.05 | 90.4 | — | — | |
| P-TuningBackbone=RoBERTa-large, Training Samples (K)=16, Evaluation Protocol=Few-shot fine-tuning2021.08 | 90.3 | — | — | |
| CMAB-GFNBackbone=GPT-J 6B, Sample size=50 Samples2026.02 | 90 | — | — | |
| NSGA-II-POModel=Qwen3-30B, Token Budget=7.5M2026.05 | 88.5 | — | — | |
| MO-CAPOModel=Qwen3-30B, Token Budget=7.5M2026.05 | 88.5 | — | — | |
| UniFew_metaShots=16/162021.07 | 87.6 | — | — | |
| NSGA-II-POModel=GPT-OSS-120B, Token Budget=7.5M2026.05 | 87.1 | — | — | |
| DP SimHash Bucketingprivacy budget (ε)=7, number of hyperplanes=24, number of hash tables=42026.05 | 87.1 | — | — | |
| DP SimHash Bucketingprivacy budget (ε)=8, number of hyperplanes=24, number of hash tables=42026.05 | 87.1 | — | — | |
| GFlowNetBackbone=GPT-J 6B, Sample size=50 Samples2026.02 | 87 | — | — | |
| GEPAModel=GPT-OSS-120B, Token Budget=7.5M2026.05 | 87 | — | — | |
| DP SimHash Bucketingprivacy budget (ε)=5, number of hyperplanes=24, number of hash tables=42026.05 | 87 | — | — | |
| MO-CAPOModel=GPT-OSS-120B, Token Budget=7.5M2026.05 | 86.9 | — | — | |
| DP SimHash Bucketingprivacy budget (ε)=1, number of hyperplanes=24, number of hash tables=42026.05 | 86.9 | — | — | |
| DP SimHash Bucketingprivacy budget (ε)=3, number of hyperplanes=24, number of hash tables=42026.05 | 86.9 | — | — | |
| DLN-2LLM=GPT-32023.06 | 85.9 | — | — | |
| CAPOModel=GPT-OSS-120B, Token Budget=7.5M2026.05 | 85.8 | — | — | |
| DP SimHash Bucketingprivacy budget (ε)=0.1, number of hyperplanes=24, number of hash tables=42026.05 | 85.2 | — | — | |
| PINFew-shot=5-shot, Policy LM=OPT-125M2024.07 | 85.1 | — | — | |
| UniFewShots=16/162021.07 | 84.6 | — | — | |
| DLN-1LLM=GPT-32023.06 | 83.2 | — | — | |
| CMAB-GFNBackbone=GPT-J 6B, Sample size=20 Samples2026.02 | 83 | — | — | |
| RLPromptFew-shot=5-shot, Policy LM=OPT-125M2024.07 | 82.2 | — | — | |
| CAPOModel=Mistral-Small-24B2025.04 | 81.67 | — | — | |
| GEPAModel=Mistral-3.2-24B, Token Budget=7.5M2026.05 | 81.3 | — | — | |
| GFlowNetBackbone=GPT-J 6B, Sample size=20 Samples2026.02 | 81 | — | — | |
| PIAST (E)variant=Ensemble2025.12 | 80.98 | — | — | |
| DLN-1LLM=WizardLM-v1.2 13B2023.06 | 79.8 | — | — | |
| AutoPromptFew-shot=5-shot2024.07 | 78.9 | — | — | |
| EvoPromptGAModel=Qwen3-30B, Token Budget=7.5M2026.05 | 78.3 | — | — | |
| CMAB-GFNBackbone=GPT-2, Sample size=50 Samples2026.02 | 78 | — | — | |
| EvoPromptGAModel=Mistral-3.2-24B, Token Budget=7.5M2026.05 | 77.9 | — | — | |
| GEPAModel=Qwen3-30B, Token Budget=7.5M2026.05 | 77.5 | — | — | |
| OPROModel=Mistral-Small-24B2025.04 | 77.07 | — | — | |
| PRL2025.12 | 76.9 | — | — | |
| PIAST2025.12 | 75.93 | — | — | |
| EvoPromptGAModel=Qwen2.5-32B2025.04 | 75.87 | — | — | |
| EvoPromptGAModel=Llama-3.3-70B2025.04 | 75.53 | — | — | |
| InitialModel=Qwen3-30B, Token Budget=7.5M2026.05 | 75.4 | — | — | |
| GFlowNetBackbone=GPT-2, Sample size=50 Samples2026.02 | 75 | — | — | |
| EvoPromptGAModel=Mistral-Small-24B2025.04 | 74.93 | — | — | |
| GA2025.12 | 74.93 | — | — | |
| GrIPSFew-shot=5-shot2024.07 | 74.8 | — | — | |
| EvoPromptGAModel=GPT-OSS-120B, Token Budget=7.5M2026.05 | 74.4 | — | — | |
| InitialModel=GPT-OSS-120B, Token Budget=7.5M2026.05 | 74.1 | — | — | |
| APE2025.12 | 73.92 | — | — | |
| DE2025.12 | 73.08 | — | — | |
| In-Context DemonstrationFew-shot=5-shot2024.07 | 73 | — | — | |
| Soft Prompt TuningFew-shot=5-shot2024.07 | 73 | — | — | |
| OPROModel=Llama-3.3-70B2025.04 | 71.33 | — | — | |
| CMAB-GFNBackbone=GPT-J 6B, Sample size=10 Samples2026.02 | 71 | — | — | |
| GFlowNetBackbone=GPT-J 6B, Sample size=10 Samples2026.02 | 71 | — | — | |
| OPROModel=Qwen2.5-32B2025.04 | 70.67 | — | — | |
| CONDACCBackbone=OPT-13B2022.12 | 70.6 | 11.9 | 50 | |
| CONDACCBackbone=GPTJ-6B2022.12 | 70.5 | 10.4 | 50.2 | |
| ALL + CALIBBackbone=GPTJ-6B2022.12 | 70.4 | 7.7 | 55.7 | |
| APO2025.12 | 69.8 | — | — |