Named Entity Recognition on Weibo (test)
70.5Overall ScoreSoftLexicon (LSTM) + BERT
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SoftLexicon (LSTM) + BERTmodeling_layer=LSTM, backbone=BERT2019.08 | 70.5 | 70.94 | 67.02 | — | — | — | |
| SoftLexicon (LSTM) + BERT2019.08 | 70.5 | 70.94 | 67.02 | — | — | — | |
| BERT + MECTpre-training=BERT, radical feature=SC2021.07 | 70.43 | — | — | — | — | — | |
| BERTpre-training=BERT2021.07 | 68.2 | — | — | — | — | — | |
| BERT + LSTM + CRFbackbone=BERT, sequence_modeling=LSTM, decoder=CRF2019.08 | 67.33 | 69.65 | 64.62 | — | — | — | |
| BERT-Taggerbackbone=BERT2019.08 | 63.8 | 65.77 | 62.05 | — | — | — | |
| MECTradical feature=SC2021.07 | 63.3 | 61.91 | 62.51 | — | — | — | |
| SoftLexicon (LSTM)modeling_layer=LSTM2019.08 | 61.42 | 59.08 | 62.22 | — | — | — | |
| SoftLexicon (LSTM)2021.07 | 61.42 | 59.08 | 62.22 | — | — | — | |
| Baseline (FLAT)2021.07 | 60.32 | — | — | — | — | — | |
| LGN2021.07 | 60.21 | 55.34 | 64.98 | — | — | — | |
| LR-CNNcitation=Gui et al., 20192019.08 | 59.92 | 57.14 | 66.67 | — | — | — | |
| LR-CNN2021.07 | 59.92 | 57.14 | 66.67 | — | — | — | |
| SoftLexicon (LSTM) + bicharmodeling_layer=LSTM, features=bichar2019.08 | 59.81 | 58.12 | 64.2 | — | — | — | |
| PLT2021.07 | 59.76 | 53.55 | 64.9 | — | — | — | |
| CAN ModelArchitecture=Convolutional Attention Network2019.04 | 59.31 | 55.38 | 62.98 | — | — | — | |
| CAN-NER2021.07 | 59.31 | 55.38 | 62.98 | — | — | — | |
| Peng and Dredze [18]Training method=Jointly train CWS task2019.04 | 58.99 | 55.28 | 62.97 | — | — | — | |
| Peng and Dredzeyear=2016, variant=*2019.08 | 58.99 | 55.28 | 62.97 | — | — | — | |
| Peng and Dredzeexternal labeled data (semi-supervised)=true2021.07 | 58.99 | 55.28 | 62.97 | — | — | — | |
| Zhang and Yang [10]Architecture=Lattice structure for lexicon integration2019.04 | 58.79 | 53.04 | 62.25 | — | — | — | |
| Lattice-LSTM2019.08 | 58.79 | 53.04 | 62.25 | — | — | — | |
| Lattice LSTM2021.07 | 58.79 | 53.04 | 62.25 | — | — | — | |
| Cao et al. [11]Approach=Adversarial transfer learning2019.04 | 58.7 | 51.34 | 57.35 | — | — | — | |
| Cao et al.2021.07 | 58.7 | 54.34 | 57.35 | — | — | — | |
| He and Sun [20]Architecture=Unified model, Data augmentation=cross-domain and semi-supervised2019.04 | 58.23 | 54.5 | 62.17 | — | — | — | |
| He and Sunyear=2017b, variant=*2019.08 | 58.23 | 54.5 | 62.17 | — | — | — | |
| He and Sunexternal labeled data (semi-supervised)=true2021.07 | 58.23 | 54.5 | 62.17 | — | — | — | |
| Char-based (LSTM) + bichar + softwordmodeling_layer=LSTM, features=bichar + softword2019.08 | 56.75 | 50.55 | 60.11 | — | — | — | |
| Peng and Dredze [16]Training method=Jointly train embeddings with NER task2019.04 | 56.05 | 51.96 | 61.05 | — | — | — | |
| Peng and Dredzeyear=20152019.08 | 56.05 | 51.96 | 61.05 | — | — | — | |
| Peng and Dredze, 20152019.08 | 56.05 | 51.96 | 61.05 | — | — | — | |
| Peng and Dredze2021.07 | 56.05 | 51.96 | 61.05 | — | — | — | |
| Char-based (LSTM) + bichar + ExSoftwordmodeling_layer=LSTM, features=bichar + ExSoftword2019.08 | 56.02 | 58.93 | 53.38 | — | — | — | |
| Baseline + CNN (CNN + BiGRU + CRF)Architecture=CNN + BiGRU + CRF2019.04 | 55.91 | 53.86 | 58.05 | — | — | — | |
| He and Sun [19]Note=Previous model baseline2019.04 | 54.82 | 50.6 | 59.32 | — | — | — | |
| He and Sunyear=2017a2019.08 | 54.82 | 50.6 | 59.32 | — | — | — | |
| He and Sun2021.07 | 54.82 | 50.6 | 59.32 | — | — | — | |
| Baseline (BiGRU + CRF)Architecture=BiGRU + CRF2019.04 | 53.8 | 49.02 | 58.8 | — | — | — | |
| Char-based (LSTM)modeling_layer=LSTM2019.08 | 52.77 | 46.11 | 55.29 | — | — | — | |
| Char-based (LSTM) + ExSoftwordmodeling_layer=LSTM, features=ExSoftword2019.08 | 52.42 | 44.65 | 55.19 | — | — | — | |
| Peng and Dredze (2016)*2018.05 | 0.5899 | 0.5528 | 0.6297 | — | — | — | |
| LatticeRepresentation=Lattice LSTM2018.05 | 0.5879 | 0.5304 | 0.6225 | — | — | — | |
| He and Sun (2017b)*2018.05 | 0.5823 | 0.545 | 0.6217 | — | — | — | |
| +bichar+softwordBase Model=Char baseline, Features=bichar+softword2018.05 | 0.5675 | 0.5055 | 0.6011 | — | — | — | |
| Peng and Dredze (2015)2018.05 | 0.5605 | 0.5196 | 0.6105 | — | — | — | |
| He and Sun (2017a)2018.05 | 0.5482 | 0.506 | 0.5932 | — | — | — | |
| Char baselineRepresentation=Character-based2018.05 | 0.5277 | 0.4611 | 0.5529 | — | — | — | |
| +char+bichar LSTMBase Model=Word baseline, Features=char+bichar2018.05 | 0.5233 | 0.434 | 0.603 | — | — | — | |
| Word baselineRepresentation=Word-based2018.05 | 0.4733 | 0.3602 | 0.5938 | — | — | — | |
| BERTModel scale=Base2021.06 | — | — | — | 67.12 | 66.88 | 67.33 | |
| ChineseBERTModel scale=Base2021.06 | — | — | — | 68.27 | 69.78 | 69.02 | |
| ChineseBERTModel scale=Large2021.06 | — | — | — | 68.75 | 72.97 | 70.8 | |
| Clean dataMethod Category=Upper Bound2023.05 | — | — | — | — | — | 68.87 | |
| CPLLType=Ours2023.05 | — | — | — | — | — | 68.23 | |
| Entity-level VotingMethod Category=Voting2023.05 | — | — | — | — | — | 59.34 | |
| Gui et al. (2019)2021.12 | — | — | — | 57.14 | 66.67 | 59.92 | |
| Li et al. (2020b)2021.12 | — | — | — | — | — | 68.55 | |
| LW lossMethod Category=PLL2023.05 | — | — | — | — | — | 64.26 | |
| Ma et al. (2020)2021.12 | — | — | — | 70.94 | 67.02 | 70.5 | |
| PRODEN-mlpMethod Category=PLL2023.05 | — | — | — | — | — | 61.85 | |
| RoBERTaModel scale=Base2021.06 | — | — | — | 68.49 | 67.81 | 68.15 | |
| RoBERTaModel scale=Large2021.06 | — | — | — | 66.74 | 70.02 | 68.35 | |
| SeqcrowdMethod Category=Annotator2023.05 | — | — | — | — | — | 41.49 | |
| Token-level VotingMethod Category=Voting2023.05 | — | — | — | — | — | 63.81 | |
| W2NER2021.12 | — | — | — | 70.84 | 73.87 | 72.32 | |
| Yan et al. (2019)2021.12 | — | — | — | — | — | 58.17 | |
| Zhang and Yang (2018)2021.12 | — | — | — | 53.04 | 62.25 | 58.79 |