Intent Detection on ATIS
99.43ID AccuracyCoBiC
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CoBiC2023.03 | 99.43 | — | |
| CharEmbed+CNN-LSTM-CRF2023.03 | 99.09 | — | |
| Federated LearningBERT used=true2023.03 | 98.28 | — | |
| SASBGCBERT used=true2023.03 | 98.21 | — | |
| CTRANBERT used=true2023.03 | 98.07 | — | |
| Co-interactive TransformerBERT used=true2023.03 | 98 | — | |
| Fine-TunedBackbone=RoBERTa-large2025.12 | 97.8 | — | |
| Bi-directional Interrelated2023.03 | 97.76 | — | |
| Joint BERTBERT used=true2023.03 | 97.5 | — | |
| Elmo+BiLSTM+CRF2023.03 | 97.42 | — | |
| ML-LMCLmanual transcripts=true2023.11 | 97.21 | — | |
| CNN-BLSTM-Aligned2023.03 | 97.17 | — | |
| ML-LMCLmanual transcripts=false2023.11 | 96.52 | — | |
| SpokenCSEmanual transcripts=true2023.11 | 95.58 | — | |
| Task MatrixBackbone=RoBERTa-large, best layer=(4,6)2025.12 | 95.5 | — | |
| Phoneme-BERTmanual transcripts=true2023.11 | 95.14 | — | |
| SpokenCSEmanual transcripts=false2023.11 | 95.1 | — | |
| Linear ProbeBackbone=all-MiniLM-L12-v2, Method (Evaluation Protocol)=Linear Probe2025.12 | 94.9 | — | |
| RoBERTamanual transcripts=true2023.11 | 94.86 | — | |
| Phoneme-BERTmanual transcripts=false2023.11 | 94.83 | — | |
| RoBERTamanual transcripts=false2023.11 | 94.53 | — | |
| SimCSEmanual transcripts=true2023.11 | 94.32 | — | |
| SimCSEmanual transcripts=false2023.11 | 94.07 | — | |
| Fine-TunedBackbone=all-MiniLM-L12-v2, Method (Evaluation Protocol)=Fine-Tuned2025.12 | 93.5 | — | |
| Task MatrixBackbone=all-MiniLM-L12-v2, Method (Evaluation Protocol)=Task Matrix, Best Layer=L52025.12 | 91.9 | — | |
| TaDSELearning Paradigm=Unsupervised, Parameters=110M2023.05 | 89.7 | — | |
| Linear ProbeBackbone=RoBERTa-large2025.12 | 89.3 | — | |
| Qwen3-0.6BLearning Paradigm=Supervised, Parameters=0.6B2023.05 | 89.14 | — | |
| Gemini-001Learning Paradigm=Supervised2023.05 | 86 | — | |
| OpenAI-smallLearning Paradigm=Supervised2023.05 | 84.88 | — | |
| OpenAI-largeLearning Paradigm=Supervised2023.05 | 84.77 | — | |
| Base w/ FT ClassifierBackbone=all-MiniLM-L12-v2, Method (Evaluation Protocol)=Base w/ FT Classifier2025.12 | 21.6 | — | |
| Attention BiRNNTraining Protocol=Independent2016.09 | — | 0.0235 | |
| Attention BiRNNTraining protocol=Joint2016.09 | — | 1.79 | |
| Attention Encoder-Decoder NNTraining Protocol=Independent2016.09 | — | 0.0202 | |
| Attention Encoder-Decoder NNTraining protocol=Joint, aligned inputs=true2016.09 | — | 1.57 | |
| BoostingTraining Protocol=Independent2016.09 | — | 0.0438 | |
| Boosting + Simplified sentencesTraining Protocol=Independent2016.09 | — | 0.0302 | |
| RecNNTraining protocol=Joint2016.09 | — | 4.6 | |
| RecNN+ViterbiTraining protocol=Joint2016.09 | — | 4.6 | |
| Recursive NNTraining Protocol=Independent2016.09 | — | 0.046 |