Sentiment Analysis on IMDB
97.44AccuracyUD+-XXL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| UD+-XXLMode=Fully-supervised, Backbone=T5-XXL encoder, Model Parameters=11B2022.11 | 97.44 | — | |
| Tay et al. (2022)Mode=Fully-supervised, Model Parameters=4x of UD (~44B)2022.11 | 97.3 | — | |
| Llama-3.3-70B-Instruct-FP8N (#rows)=25,000, K (#candidate labels)=2, Decoding Strategy=greedy2026.05 | 95.1 | — | |
| gpt-oss-120BN (#rows)=25,000, K (#candidate labels)=2, Decoding Strategy=prefill-only logit scoring2026.05 | 94.1 | — | |
| ColD-Fusion2022.12 | 94.01 | — | |
| IMO-BARTBackbone=BART2024.04 | 93.97 | — | |
| Finetune2022.12 | 93.86 | — | |
| Multitask2022.12 | 93.82 | — | |
| BERT2026.04 | 93.46 | — | |
| Fine-tuningn=256, Update model parameters=true2022.12 | 92.1 | — | |
| DecTn=256, Update model parameters=false2022.12 | 92.1 | — | |
| LLM-Guided TM2026.04 | 92.1 | — | |
| MUPPET2022.12 | 91.74 | — | |
| IMO-BART lastBackbone=BART, Masking=last layer only2024.04 | 91.71 | — | |
| DecTn=64, Update model parameters=false2022.12 | 91.3 | — | |
| DecTn (Shots)=12022.12 | 91.2 | — | |
| CHATGPTModel=gpt-3.5-turbo, Evaluation protocol=Few-shot in-context learning2024.04 | 91.08 | — | |
| DecTn (Shots)=162022.12 | 91 | — | |
| TMEmbedding=GloVe2026.04 | 90.88 | — | |
| TM2026.04 | 90.62 | — | |
| ALPACA-7BBackbone=LLaMA 7B, Tuning=fine-tuned2024.04 | 90.14 | — | |
| BARTTuning=fine-tuned2024.04 | 89.91 | — | |
| BBTn (Shots)=42022.12 | 89.8 | — | |
| ALPACA-7B-LORABackbone=LLaMA 7B, Tuning=LoRA fine-tuned2024.04 | 89.8 | — | |
| BERTTuning=fine-tuned2024.04 | 89.77 | — | |
| DecTn (Shots)=42022.12 | 89.6 | — | |
| Promptn (Shots)=02022.12 | 89.4 | — | |
| BBTv2n (Shots)=42022.12 | 89.4 | — | |
| PDAApproach=Prompt learning2024.04 | 89.35 | — | |
| BBTn (Shots)=162022.12 | 89.3 | — | |
| GNN2025.07 | 89.19 | — | |
| BBTn (Shots)=12022.12 | 89 | — | |
| BBTv2n (Shots)=12022.12 | 89 | — | |
| BBTv2n (Shots)=162022.12 | 88.6 | — | |
| BERT-PGBData augmentation=PGB2024.04 | 88.4 | — | |
| GraphTMDepth=22025.07 | 88.15 | — | |
| REGENClassifier=DistillBERT2023.05 | 87.84 | — | |
| BERT-UDAData augmentation=UDA2024.04 | 87.76 | — | |
| BERT-EDAData augmentation=EDA2024.04 | 87.73 | — | |
| RLPromptn (Shots)=162022.12 | 87.6 | — | |
| TS-ELSTMArchitecture Type=2-state enhanced LSTM (TS-ELSTM) + Emotional intelligence, Sequence Length=18252025.12 | 87.24 | — | |
| GraphTMDepth=12025.07 | 86.43 | — | |
| Fine-tuningn=64, Update model parameters=true2022.12 | 86.3 | — | |
| PromptBoostingn (Shots)=162022.12 | 86.2 | — | |
| PADAApproach=Prompt learning2024.04 | 85.73 | — | |
| Baseline#seeds=4, Backbone=DistilBERT, Training Protocol=Pre-trained2026.04 | 85.69 | — | |
| BoW TFIDF2026.04 | 85.3 | — | |
| LottaLoRARank=8, #seeds=4, % trainable=0.48%, Backbone=DistilBERT, Training Protocol=Pre-trained2026.04 | 85.12 | — | |
| StdTM2025.07 | 84.6 | — | |
| SuperGenClassifier=DistillBERT2023.05 | 84.58 | — | |
| ProGenClassifier=DistillBERT2023.05 | 84.12 | — | |
| PromptBoostingn (Shots)=42022.12 | 83 | — | |
| PromptBoostingn (Shots)=12022.12 | 82.4 | — | |
| RNN/LSTMEmbedding=GloVe2026.04 | 80.72 | — | |
| ICLn (Shots)=162022.12 | 80.6 | — | |
| ZeroGenClassifier=DistillBERT2023.05 | 80.41 | — | |
| ICLn (Shots)=42022.12 | 80.4 | — | |
| char-CNN2026.04 | 80.35 | — | |
| RNN/LSTM2026.04 | 78.82 | — | |
| MiningClassifier=DistillBERT2023.05 | 77.36 | — | |
| PromptingClassifier=DistillBERT2023.05 | 77.31 | — | |
| IMO-BART B2TBackbone=BART, Search strategy=bottom-up layer-wise search2024.04 | 75.86 | — | |
| RLPromptn (Shots)=42022.12 | 75.8 | — | |
| IMO-BART w/o sqBackbone=BART, Search strategy=simultaneous search2024.04 | 74.88 | — | |
| Zero-shot PromptingEst. Cost=3.42024.06 | 73.7 | — | |
| TranskimmerBackbone=BERT-base, Padding=No2022.05 | 70.1 | 2.51 | |
| TR-BERTBackbone=BERT-base, Padding=No2022.05 | 70 | 2.19 | |
| BERT_baseBackbone=BERT-base, Padding=No2022.05 | 69.9 | 1 | |
| POWER-BERTBackbone=BERT-base, Padding=Sequence2022.05 | 67.4 | 2.75 | |
| Alchemist with GPT-3.5Est. Cost=0.0042024.06 | 66.2 | — | |
| ICLn (Shots)=12022.12 | 65.6 | — | |
| RLPromptn (Shots)=12022.12 | 65 | — | |
| Minimal Gated Unit (MGU)Architecture Type=Reduced-gate RNN, Parameters (approx.)=20–22M, Sequence Length=1282025.12 | 62.6 | — |