Sentiment Classification on Yelp (test)
96.4AccuracyICL-gold
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ICL-goldModel=GPT-J, Prompting=Direct, Training data access=true2022.12 | 96.4 | — | — | |
| Z-ICLModel=GPT-J, Prompting=Direct, Training data access=false2022.12 | 96 | — | — | |
| ICL-randomModel=GPT-J, Prompting=Direct, Training data access=true2022.12 | 94.5 | — | — | |
| NLI-STLearning Paradigm=Reference, Use auxiliary labeled data=true, Use task-specific corpus=true2023.05 | 94.3 | — | — | |
| SuperGenLearning Paradigm=Zero-shot Learning via Generating Task-specific Datasets, Standard Deviation=0.62023.05 | 93.6 | — | — | |
| REGENLearning Paradigm=Zero-shot Learning via Generating Task-specific Datasets, Standard Deviation=0.52023.05 | 93 | — | — | |
| KPTLearning Paradigm=Reference, Use task-specific corpus=true, Use additional knowledge base=true2023.05 | 92.8 | — | — | |
| Mining (Re-implementation)Learning Paradigm=Zero-shot Learning via Generating Task-specific Datasets, Concurrent Work=true, Fair Comparison Setup=true, Standard Deviation=0.52023.05 | 92.3 | — | — | |
| No-demosModel=GPT-J, Prompting=Direct, Training data access=false2022.12 | 92.3 | — | — | |
| Z-ICLModel=GPT-NeoX, Prompting=Direct, Training data access=false2022.12 | 92.2 | — | — | |
| MiningLearning Paradigm=Zero-shot Learning via Generating Task-specific Datasets, Concurrent Work=true2023.05 | 92 | — | — | |
| Naive Z-ICLModel=GPT-J, Prompting=Direct, Training data access=false2022.12 | 91.7 | — | — | |
| ICL-goldModel=GPT-NeoX, Prompting=Direct, Training data access=true2022.12 | 91.7 | — | — | |
| Random inputsModel=GPT-J, Prompting=Direct, Training data access=false2022.12 | 91.5 | — | — | |
| ICL-randomModel=GPT-J, Prompting=Channel, Training data access=true2022.12 | 91.1 | — | — | |
| ICL-goldModel=GPT-J, Prompting=Channel, Training data access=true2022.12 | 91 | — | — | |
| AttrPromptClassifier=DistilBERT, Training Protocol=Simple fine-tuning with standard cross-entropy loss2023.06 | 90.6 | — | — | |
| X-ClassLearning Paradigm=Reference, Use task-specific corpus=true2023.05 | 90 | — | — | |
| SuperGenClassifier=DistilBERT, Training Protocol=Dedicated training techniques2023.06 | 89.88 | — | — | |
| ProGenClassifier=DistilBERT, Training Protocol=Dedicated training techniques2023.06 | 89.39 | — | — | |
| SunGenClassifier=DistilBERT, Training Protocol=Dedicated training techniques2023.06 | 89.19 | — | — | |
| Naive Z-ICLModel=GPT-NeoX, Prompting=Direct, Training data access=false2022.12 | 89 | — | — | |
| ReGenClassifier=DistilBERT, Training Protocol=Dedicated training techniques2023.06 | 89 | — | — | |
| ICL-randomModel=GPT-NeoX, Prompting=Direct, Training data access=true2022.12 | 88.5 | — | — | |
| Z-ICLModel=GPT-J, Prompting=Channel, Training data access=false2022.12 | 88.4 | — | — | |
| SimPromptClassifier=DistilBERT, Training Protocol=Simple fine-tuning with standard cross-entropy loss2023.06 | 88.39 | — | — | |
| No-demosModel=GPT-J, Prompting=Channel, Training data access=false2022.12 | 88 | — | — | |
| ICL-randomModel=GPT-NeoX, Prompting=Channel, Training data access=true2022.12 | 88 | — | — | |
| ZeroGenClassifier=DistilBERT, Training Protocol=Dedicated training techniques2023.06 | 87.84 | — | — | |
| LOTClassLearning Paradigm=Reference, Use task-specific corpus=true2023.05 | 87.6 | — | — | |
| Z-ICLModel=GPT-NeoX, Prompting=Channel, Training data access=false2022.12 | 87 | — | — | |
| ICL-goldModel=GPT-NeoX, Prompting=Channel, Training data access=true2022.12 | 86.8 | — | — | |
| KNN-PromptLearning Paradigm=Zero-shot Learning via Direct Inferencing2023.05 | 85.7 | — | — | |
| Random inputsModel=GPT-NeoX, Prompting=Direct, Training data access=false2022.12 | 85 | — | — | |
| Random inputsModel=GPT-J, Prompting=Channel, Training data access=false2022.12 | 84.2 | — | — | |
| Naive Z-ICLModel=GPT-J, Prompting=Channel, Training data access=false2022.12 | 81.4 | — | — | |
| Naive Z-ICLModel=GPT-NeoX, Prompting=Channel, Training data access=false2022.12 | 79.1 | — | — | |
| GPT-3Learning Paradigm=Zero-shot Learning via Direct Inferencing, Billion-scale PLM=true2023.05 | 78.5 | — | — | |
| PromptLearning Paradigm=Zero-shot Learning via Direct Inferencing2023.05 | 78.1 | — | — | |
| Zero-shot PromptingClassifier=DistilBERT, Training Protocol=Dedicated training techniques2023.06 | 78.1 | — | — | |
| TE-NLI (Best)Learning Paradigm=Labeled data usage, Use auxiliary labeled data=true2023.05 | 73.5 | — | — | |
| Random inputsModel=GPT-NeoX, Prompting=Channel, Training data access=false2022.12 | 73 | — | — | |
| No-demosModel=GPT-NeoX, Prompting=Direct, Training data access=false2022.12 | 72.2 | — | — | |
| NSP-BERTLearning Paradigm=Zero-shot Learning via Direct Inferencing2023.05 | 66.3 | — | — | |
| No-demosModel=GPT-NeoX, Prompting=Channel, Training data access=false2022.12 | 57 | — | — | |
| MajorityTraining data access=false2022.12 | 50 | — | — | |
| AttrPromptBase Model=Llama3-8B-Instruct2025.04 | — | — | 86.4 | |
| Golden data2025.04 | — | — | 89.6 | |
| LHTR2020.03 | — | 0.152 | — | |
| LHTR12020.03 | — | 0.155 | — | |
| MoPBase Model=Llama3-8B-Instruct2025.04 | — | — | 86.7 | |
| NN model2020.03 | — | 0.111 | — | |
| PICLeBase Model=Llama3-8B-Instruct, prompting=in-context2025.04 | — | — | 73.8 | |
| ProGenBase Model=Llama3-8B-Instruct2025.04 | — | — | 84.3 | |
| Proposed ModelComponents=NN model (bulk) + LHTR (extreme)2020.03 | — | 0.103 | — | |
| ZeroGenBase Model=Llama3-8B-Instruct, prompting=zero-shot2025.04 | — | — | 86 |