Text Classification on Yahoo! Answers (test)
74.8Clean AccuracyEXAM
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| EXAMModel Type=Word-based2018.11 | 74.8 | — | — | — | — | — | |
| StandardArchitecture=BiLSTM2025.07 | 74.7 | 12.2 | 9.6 | 6.5 | — | — | |
| StandardModel Architecture=BiLSTM2026.06 | 74.7 | 12.2 | 9.6 | 6.5 | — | — | |
| OoMMixTrain size=25K2021.05 | 74.13 | — | — | — | — | — | |
| MixText†Train size=25K, unlabeled data=50K2021.05 | 74.1 | — | — | — | — | — | |
| TMixTrain size=25K2021.05 | 73.84 | — | — | — | — | — | |
| MLCClassifier=Pre-trained BERT-base, Number of clean labels=1000, Noise types=2, Noise levels=10, k (SGD steps)=52019.11 | 73.72 | — | — | — | — | — | |
| W.C RegionEmbModel Type=Word-based2018.11 | 73.7 | — | — | — | — | — | |
| OriginalTrain size=25K2021.05 | 73.68 | — | — | — | — | — | |
| ATFLArchitecture=BiLSTM2025.07 | 73.6 | 61.7 | 60.8 | 60.3 | — | — | |
| ATFLModel Architecture=BiLSTM2026.06 | 73.6 | 61.7 | 60.8 | 60.3 | — | — | |
| SWEM-concat2018.05 | 73.53 | — | — | — | — | — | |
| mixup-transformerTrain size=25K2021.05 | 73.52 | — | — | — | — | — | |
| TMix†Train size=25K2021.05 | 73.5 | — | — | — | — | — | |
| SWEM-hier2018.05 | 73.48 | — | — | — | — | — | |
| Deep CNNlayers=292018.05 | 73.43 | — | — | — | — | — | |
| VDCNNModel Type=Character-based2018.11 | 73.4 | — | — | — | — | — | |
| SWEM-aver2018.05 | 73.14 | — | — | — | — | — | |
| GBMArchitecture=BiLSTM2025.07 | 73 | 67 | 67 | 68.6 | — | — | |
| GBMModel Architecture=BiLSTM2026.06 | 73 | 67 | 67 | 68.6 | — | — | |
| GBMArchitecture=CNN2025.07 | 72.8 | 66.2 | 66.1 | 67.3 | — | — | |
| GBMModel Architecture=CNN2026.06 | 72.8 | 66.2 | 66.1 | 67.3 | — | — | |
| SWEM-max2018.05 | 72.66 | — | — | — | — | — | |
| StandardArchitecture=CNN2025.07 | 72.6 | 6.8 | 7.2 | 4.9 | — | — | |
| StandardModel Architecture=CNN2026.06 | 72.6 | 6.8 | 7.2 | 4.9 | — | — | |
| ATFLArchitecture=CNN2025.07 | 72.5 | 62.5 | 63.1 | 62.5 | — | — | |
| ATFLModel Architecture=CNN2026.06 | 72.5 | 62.5 | 63.1 | 62.5 | — | — | |
| SEMArchitecture=BiLSTM2025.07 | 72.3 | 57 | 56.1 | 55.4 | — | — | |
| fastText (bigram)features=bigram2018.05 | 72.3 | — | — | — | — | — | |
| Bigram-FastTextModel Type=Word-based2018.11 | 72.3 | — | — | — | — | — | |
| SEMModel Architecture=BiLSTM2026.06 | 72.3 | 57 | 56.1 | 55.4 | — | — | |
| fastText2018.05 | 72 | — | — | — | — | — | |
| Char-CRNNModel Type=Character-based2018.11 | 71.7 | — | — | — | — | — | |
| MixText†Train size=2K, unlabeled data=50K2021.05 | 71.3 | — | — | — | — | — | |
| Char-CNNModel Type=Character-based2018.11 | 71.2 | — | — | — | — | — | |
| OoMMixTrain size=2K2021.05 | 71.08 | — | — | — | — | — | |
| Large word CNNModel Type=Word-based2018.11 | 71 | — | — | — | — | — | |
| Large word CNN2018.05 | 70.94 | — | — | — | — | — | |
| LSTM2018.05 | 70.84 | — | — | — | — | — | |
| LSTMModel Type=Word-based2018.11 | 70.8 | — | — | — | — | — | |
| ASCCArchitecture=BiLSTM2025.07 | 70.7 | 61.7 | 62.3 | 61.9 | — | — | |
| ASCCModel Architecture=BiLSTM2026.06 | 70.7 | 61.7 | 62.3 | 61.9 | — | — | |
| TMixTrain size=2K2021.05 | 70.68 | — | — | — | — | — | |
| Unified PPTTuning Mode=Prompt tuning, Parameters=410K2021.09 | 70.5 | — | — | — | — | — | |
| OriginalTrain size=2K2021.05 | 70.41 | — | — | — | — | — | |
| mixup-transformerTrain size=2K2021.05 | 70.29 | — | — | — | — | — | |
| SEMArchitecture=CNN2025.07 | 70.1 | 53.8 | 52.4 | 51.9 | — | — | |
| SEMModel Architecture=CNN2026.06 | 70.1 | 53.8 | 52.4 | 51.9 | — | — | |
| Small word CNNModel Type=Word-based2018.11 | 70 | — | — | — | — | — | |
| Small word CNN2018.05 | 69.98 | — | — | — | — | — | |
| TMix†Train size=2K2021.05 | 69.8 | — | — | — | — | — | |
| NonlinearMixTrain size=25K2021.05 | 69.31 | — | — | — | — | — | |
| NonlinearMixTrain size=2K2021.05 | 69.17 | — | — | — | — | — | |
| ASCCArchitecture=CNN2025.07 | 69 | 58.4 | 59.6 | 58.5 | — | — | |
| ASCCModel Architecture=CNN2026.06 | 69 | 58.4 | 59.6 | 58.5 | — | — | |
| BoWModel Type=Feature Engineering2018.11 | 68.9 | — | — | — | — | — | |
| N-gramsModel Type=Feature Engineering2018.11 | 68.5 | — | — | — | — | — | |
| N-grams TFIDFModel Type=Feature Engineering2018.11 | 68.5 | — | — | — | — | — | |
| GLCClassifier=Pre-trained BERT-base, Number of clean labels=1000, Noise types=2, Noise levels=10, k (SGD steps)=52019.11 | 68.03 | — | — | — | — | — | |
| OoMMixTrain size=0.5K2021.05 | 67.95 | — | — | — | — | — | |
| mixup-transformerTrain size=0.5K2021.05 | 67.62 | — | — | — | — | — | |
| TMixTrain size=0.5K2021.05 | 67.57 | — | — | — | — | — | |
| NonlinearMixTrain size=0.5K2021.05 | 67.56 | — | — | — | — | — | |
| SCNN-VAE-Semi+initNumber of labeled samples=2000, Encoder Initialization=Language Model2017.02 | 67.4 | — | — | — | — | — | |
| OriginalTrain size=0.5K2021.05 | 67.24 | — | — | — | — | — | |
| ProtoVerb+K (shots)=8, Tuning Strategy=Prompt-based tuning2022.03 | 66.61 | — | — | — | — | — | |
| SCNN-VAE-Semi+initNumber of labeled samples=1000, Encoder Initialization=Language Model2017.02 | 66.6 | — | — | — | — | — | |
| S-GBTModel Architecture=BiLSTM2026.06 | 66.4 | 88 | 87.5 | 89.8 | — | — | |
| ProtoVerb+K (shots)=4, Tuning Strategy=Prompt-based tuning2022.03 | 66.14 | — | — | — | — | — | |
| SCNN-VAE-SemiNumber of labeled samples=10002017.02 | 66 | — | — | — | — | — | |
| SCNN-VAE-SemiNumber of labeled samples=20002017.02 | 65.8 | — | — | — | — | — | |
| ProtoVerb+K (shots)=16, Tuning Strategy=Prompt-based tuning2022.03 | 65.65 | — | — | — | — | — | |
| LM-LSTMNumber of labeled samples=2000, Initialization=Language model2017.02 | 65.6 | — | — | — | — | — | |
| SCNN-VAE-SemiNumber of labeled samples=5002017.02 | 65.6 | — | — | — | — | — | |
| SCNN-VAE-Semi+initNumber of labeled samples=500, Encoder Initialization=Language Model2017.02 | 65.4 | — | — | — | — | — | |
| SearchVerb+K (shots)=8, Tuning Strategy=Prompt-based tuning2022.03 | 65.32 | — | — | — | — | — | |
| ProtoVerbshots (K)=16, task-specific knowledge=false2022.03 | 64.35 | — | — | — | — | — | |
| Fine-tuningK (shots)=16, Tuning Strategy=Standard fine-tuning2022.03 | 64.27 | — | — | — | — | — | |
| ManualVerbK (shots)=8, Tuning Strategy=Prompt-based tuning2022.03 | 64.12 | — | — | — | — | — | |
| ManualVerbshots (K)=8, task-specific knowledge=true2022.03 | 64.12 | — | — | — | — | — | |
| FTTuning Mode=Full-model tuning, Parameters=11B2021.09 | 64.1 | — | — | — | — | — | |
| LM-LSTMNumber of labeled samples=1000, Initialization=Language model2017.02 | 63.9 | — | — | — | — | — | |
| SCNN-VAE-Semi+initNumber of labeled samples=100, Encoder Initialization=Language Model2017.02 | 63.8 | — | — | — | — | — | |
| SoftVerb+K (shots)=8, Tuning Strategy=Prompt-based tuning2022.03 | 63.48 | — | — | — | — | — | |
| S-GBTModel Architecture=CNN2026.06 | 63.3 | 88.6 | 89 | 90.7 | — | — | |
| IBPArchitecture=CNN2025.07 | 63.1 | 54.9 | 54.9 | 54.8 | — | — | |
| IBPModel Architecture=CNN2026.06 | 63.1 | 54.9 | 54.9 | 54.8 | — | — | |
| PTTuning Mode=Prompt tuning, Parameters=410K2021.09 | 62 | — | — | — | — | — | |
| ManualVerbK (shots)=4, Tuning Strategy=Prompt-based tuning2022.03 | 61.41 | — | — | — | — | — | |
| ManualVerbshots (K)=4, task-specific knowledge=true2022.03 | 61.41 | — | — | — | — | — | |
| ProtoVerbshots (K)=8, task-specific knowledge=false2022.03 | 61.4 | — | — | — | — | — | |
| LM-LSTMNumber of labeled samples=500, Initialization=Language model2017.02 | 61.3 | — | — | — | — | — | |
| ProtoVerb+K (shots)=16, Tuning Strategy=Without tuning (applied to un-tuned PLMs)2022.03 | 60.89 | — | — | — | — | — | |
| PT (MC)Tuning Mode=Prompt tuning, Format=Multiple-choice classification format without prompt pre-training2021.09 | 60.8 | — | — | — | — | — | |
| Bag-of-means2018.05 | 60.55 | — | — | — | — | — | |
| MW-NetClassifier=Pre-trained BERT-base, Number of clean labels=1000, Noise types=2, Noise levels=10, k (SGD steps)=52019.11 | 60.18 | — | — | — | — | — | |
| SearchVerbshots (K)=16, task-specific knowledge=false2022.03 | 59.66 | — | — | — | — | — | |
| ProtoVerb+K (shots)=8, Tuning Strategy=Without tuning (applied to un-tuned PLMs)2022.03 | 59.42 | — | — | — | — | — | |
| SearchVerb+K (shots)=16, Tuning Strategy=Prompt-based tuning2022.03 | 59.34 | — | — | — | — | — | |
| ProtoVerb+K (shots)=2, Tuning Strategy=Prompt-based tuning2022.03 | 59.33 | — | — | — | — | — |