Relation Extraction on TACRED (test)
75.5F1 ScoreUNIST_LARGE
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| UNIST_LARGEPLM Scale=Large2022.05 | 75.5 | 78 | 73.1 | — | — | |
| RECENT_SpanBERTbase_model=SpanBERT2021.05 | 75.2 | 90.9 | 64.2 | — | — | |
| SP_relation, ALBERTPre-trained Model=ALBERT2020.10 | 74.8 | 74.6 | 75.2 | — | — | |
| SPPLM Scale=Large2022.05 | 74.8 | 74.6 | 75.2 | — | — | |
| IBREPLM Scale=Large2022.05 | 74.6 | — | — | — | — | |
| UNIST_BASEPLM Scale=Base2022.05 | 74.3 | 73.6 | 75 | — | — | |
| NLI_DeBERTaTraining setting=Full training2021.09 | 73.9 | 72.5 | 75.3 | — | — | |
| SP_token, ALBERTPre-trained Model=ALBERT2020.10 | 73.4 | 72.2 | 74.6 | — | — | |
| BERT-CRPLM Scale=Large2022.05 | 73 | — | — | — | — | |
| LUKE2020.10 | 72.7 | 70.4 | 75.1 | — | — | |
| LUKE2021.05 | 72.7 | 70.4 | 75.1 | — | — | |
| LUKEPLM Scale=Large2022.05 | 72.7 | 70.4 | 75.1 | — | — | |
| LUKESemantic Information=false, Correlation Information=false2022.12 | 72.7 | — | — | — | — | |
| LUKEExtra Data=w/, Learning Paradigm=Fine-tuning2021.04 | 72.7 | — | — | — | — | |
| LUKETraining setting=Full training2021.09 | 72.7 | 70.4 | 75.1 | — | — | |
| SP_question, ALBERTPre-trained Model=ALBERT2020.10 | 72.6 | 73.3 | 71.8 | — | — | |
| KEPLER_ROBERTa+KGPre-trained Model=RoBERTa-Large, External Knowledge=Knowledge Graph (Wikipedia/Wikidata)2020.10 | 72.5 | 72.8 | 72.2 | — | — | |
| PTRExtra Data=w/o, Backbone=RoBERTa, Learning Paradigm=Prompt-tuning2021.04 | 72.4 | — | — | — | — | |
| KnowPromptExtra Data=w/o, Backbone=RoBERTa, Learning Paradigm=Prompt-tuning2021.04 | 72.4 | — | — | — | — | |
| K-Adapter2021.05 | 72.04 | 70.14 | 74.04 | — | — | |
| K-Adapter2020.10 | 72 | 68.9 | 75.4 | — | — | |
| KEPLER-WikiBackbone=RoBERTa-Base2019.11 | 72 | 71.5 | 72.5 | — | — | |
| KEPLER2021.05 | 72 | 71.5 | 72.5 | — | — | |
| K-AdapterPLM Scale=Large2022.05 | 72 | 70.1 | 74 | — | — | |
| K-AdapterTraining setting=Full training2021.09 | 72 | 70.1 | 74 | — | — | |
| TANLPLM Scale=Base2022.05 | 71.9 | — | — | — | — | |
| TANL(T5)Semantic Information=true, Correlation Information=true, Backbone=T52022.12 | 71.9 | — | — | — | — | |
| SP_question, BERTPre-trained Model=BERT-Large2020.10 | 71.8 | 71.1 | 72.6 | — | — | |
| KEPLER2020.10 | 71.7 | 70.4 | 73 | — | — | |
| BERT_EM + MTBBackbone=BERT-Large, Pre-training=Matching the Blanks (MTB)2019.06 | 71.5 | — | — | — | — | |
| Soares et al. (2019)LM=BERT_LARGE, MTB pretraining=true2019.09 | 71.5 | — | — | — | — | |
| KnowBert-W+WLM=BERT_BASE2019.09 | 71.5 | 71.6 | 71.4 | — | — | |
| BERTEM + MTBpre-training=matching the blanks2019.07 | 71.5 | — | — | — | — | |
| MTB2020.10 | 71.5 | — | — | — | — | |
| KnowBERT2020.10 | 71.5 | 71.6 | 71.4 | — | — | |
| MTB (BERT_LARGE)Backbone=BERT-Large2019.11 | 71.5 | — | — | — | — | |
| KEPLER-W+WBackbone=RoBERTa-Base2019.11 | 71.5 | 71.1 | 72 | — | — | |
| DG-SpanBERTmodel_size=large2020.04 | 71.5 | 71.4 | 71.6 | — | — | |
| MTB2021.05 | 71.5 | — | — | — | — | |
| KnowBert2021.05 | 71.5 | 71.6 | 71.1 | — | — | |
| MTBPLM Scale=Large2022.05 | 71.5 | — | — | — | — | |
| MTBSemantic Information=false, Correlation Information=false2022.12 | 71.5 | — | — | — | — | |
| KNOWBERTExtra Data=w/, Learning Paradigm=Fine-tuning2021.04 | 71.5 | — | — | — | — | |
| GDPNETExtra Data=w/o, Learning Paradigm=Fine-tuning2021.04 | 71.5 | — | — | — | — | |
| KEPLER-CondBackbone=RoBERTa-Base2019.11 | 71.4 | 72.1 | 70.7 | — | — | |
| RoBERTa2020.10 | 71.3 | 70.2 | 72.4 | — | — | |
| KEPLER-WordNetBackbone=RoBERTa-Base2019.11 | 71.3 | 71.4 | 71.3 | — | — | |
| RoBERTaTraining setting=Full training2021.09 | 71.3 | 70.2 | 72.4 | — | — | |
| SP_relation, BERTPre-trained Model=BERT-Large2020.10 | 71.2 | 67 | 76 | — | — | |
| RELASemantic Information=true, Correlation Information=true, Backbone=BART-large, Approach=BART-Synonym2022.12 | 71.2 | — | — | — | — | |
| KEPLER-RelBackbone=RoBERTa-Base2019.11 | 71.1 | 71.3 | 70.9 | — | — | |
| NLI_RoBERTaTraining setting=Full training2021.09 | 71 | 71.6 | 70.4 | — | — | |
| KnowBert_ROBERTABackbone=RoBERTa-Base, implementation=reproduced2019.11 | 70.9 | 71.9 | 69.9 | — | — | |
| RECENT_GCNbase_model=GCN2021.05 | 70.9 | 88.3 | 59.3 | — | — | |
| SpanBERT-ALT2021.05 | 70.9 | 69 | 73 | — | — | |
| SpanBERT2019.07 | 70.8 | 70.8 | 70.9 | — | — | |
| SpanBERT2020.10 | 70.8 | 70.8 | 70.9 | — | — | |
| SpanBertmodel_size=large2020.04 | 70.8 | 71.2 | 70.4 | — | — | |
| SpanBERT2021.05 | 70.8 | 70.8 | 70.9 | — | — | |
| SpanBERTPLM Scale=Large2022.05 | 70.8 | 70.8 | 70.9 | — | — | |
| SPANBERTExtra Data=w/, Learning Paradigm=Fine-tuning2021.04 | 70.8 | — | — | — | — | |
| SpanBERTTraining setting=Full training2021.09 | 70.8 | 70.8 | 70.9 | — | — | |
| RoBERTaBackbone=RoBERTa-Base2019.11 | 70.7 | 70.4 | 71.1 | — | — | |
| ERNIE_ROBERTABackbone=RoBERTa-Base, implementation=reproduced2019.11 | 70.7 | 73.5 | 68 | — | — | |
| KEPLER-OnlyDescBackbone=RoBERTa-Base2019.11 | 70.7 | 72.3 | 69.1 | — | — | |
| REBELSemantic Information=true, Correlation Information=false2022.12 | 70.7 | — | — | — | — | |
| Google Bertmodel_size=large2020.04 | 70.6 | 69.2 | 72 | — | — | |
| IRE-RoBERTaSemantic Information=false, Correlation Information=false, Backbone=RoBERTa-large2022.12 | 70.5 | — | — | — | — | |
| BART-RESemantic Information=true, Correlation Information=true, Backbone=BART-large2022.12 | 70.4 | — | — | — | — | |
| Our RoBERTaBackbone=RoBERTa-Base, status=authors' baseline2019.11 | 70.2 | 70.8 | 69.6 | — | — | |
| DG-SpanBERTmodel_size=base2020.04 | 70.2 | 68.3 | 72.1 | — | — | |
| BERT_EMBackbone=BERT-Large2019.06 | 70.1 | — | — | — | — | |
| Soares et al. (2019)LM=BERT_LARGE2019.09 | 70.1 | — | — | — | — | |
| BERTEM2019.07 | 70.1 | — | — | — | — | |
| Our BERT-1seqtraining_objective=single-sequence2019.07 | 70.1 | 72.4 | 67.9 | — | — | |
| BERT_LARGEBackbone=BERT-Large2019.11 | 70.1 | — | — | — | — | |
| RC_MTB, BERTPre-trained Model=BERT-Large2020.10 | 70.1 | — | — | — | — | |
| BART-DCSemantic Information=true, Correlation Information=false, Backbone=BART-large2022.12 | 70.1 | — | — | — | — | |
| MTBExtra Data=w/, Learning Paradigm=Fine-tuning2021.04 | 70.1 | — | — | — | — | |
| SP_token, BERTPre-trained Model=BERT-Large2020.10 | 70 | 63.3 | 78.4 | — | — | |
| IRE-BERTSemantic Information=false, Correlation Information=false, Backbone=BERT-large2022.12 | 69.7 | — | — | — | — | |
| BART-DSSemantic Information=true, Correlation Information=false, Backbone=BART-large2022.12 | 69.3 | — | — | — | — | |
| BART-DBSemantic Information=false, Correlation Information=true, Backbone=BART-large2022.12 | 69.1 | — | — | — | — | |
| C-AGGCNModel Type=Dependency-based, Tree Context=full trees2019.06 | 69 | 73.1 | 64.2 | — | — | |
| C-AGGCN2021.05 | 69 | 71.8 | 66.4 | — | — | |
| MTBBackbone=BERT-Base2019.11 | 68.8 | 69.7 | 67.9 | — | — | |
| FINE-TUNINGExtra Data=w/o, Backbone=RoBERTa, Learning Paradigm=Fine-tuning2021.04 | 68.7 | — | — | — | — | |
| KnowBert_BERTBackbone=BERT-Base2019.11 | 68.5 | 73.5 | 64.1 | — | — | |
| C-GCN + PA-LSTMModel Type=Interpolated Model2018.09 | 68.2 | 71.3 | 65.4 | — | — | |
| Zhang et al. (2018)ensemble=true2019.04 | 68.2 | 71.3 | 65.4 | — | — | |
| SOTA2019.06 | 68.2 | — | — | — | — | |
| SpanBertmodel_size=base2020.04 | 68.1 | 67.6 | 68.6 | — | — | |
| ERNIELM=BERT_BASE2019.09 | 68 | 70 | 66.1 | — | — | |
| ERNIE2020.10 | 68 | 70 | 66.1 | — | — | |
| ERNIE_BERTBackbone=BERT-Base2019.11 | 68 | 70 | 66.1 | — | — | |
| ERNIEmode=fine-tuned2019.05 | 67.97 | 69.97 | 66.08 | — | — | |
| Google Bertmodel_size=base2020.04 | 67.9 | 68.1 | 67.7 | — | — | |
| BERT-LSTM-baseBackbone=BERT-base-cased2019.04 | 67.8 | 73.3 | 63.1 | — | — | |
| Shi and Lin (2019)LM=BERT_BASE2019.09 | 67.8 | 73.3 | 63.1 | — | — | |
| C-AGGCN2020.04 | 67.8 | 69.6 | 66 | — | — |