Emotion Recognition in Conversation on MELD
70.44Weighted Avg F1PRC-Emo
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| PRC-EmoBackbone=Qwen2.5/3, External FREE=NO, Training Protocol=Dataset-Specific Setting2026.04 | 70.44 | — | — | 71.5 | |
| SELF-EMOBackbone=Qwen3-8B, External FREE=YES, Training Protocol=Dataset-Unified Setting2026.04 | 70.3 | — | — | 71.92 | |
| BiosERCBackbone=LLaMA2-13B, External FREE=NO, Training Protocol=Dataset-Specific Setting2026.04 | 69.83 | — | — | — | |
| Llama2Parameters=12.5M(7B), Extra Knowledge=✗, Model type=LLM-based, Method Setting=LoRA + InstructERC2023.09 | 69.15 | — | — | — | |
| InstructERCBackbone=LLaMA2-7B, External FREE=NO, Training Protocol=Dataset-Specific Setting2026.04 | 69.15 | — | — | — | |
| TODKATAverage of ten runs=true2021.06 | 68.23 | — | 64.75 | — | |
| ML-SAN2026.04 | 67.73 | — | — | — | |
| LlamaParameters=12.5M(7B), Extra Knowledge=✗, Model type=LLM-based, Method Setting=LoRA + InstructERC2023.09 | 67.62 | — | — | — | |
| HyperEmo-ChatGLM3-6BBackbone=ChatGLM3-6B2026.05 | 67.54 | — | — | 68.88 | |
| SELF-EMOBackbone=Qwen3-4B, External FREE=YES, Training Protocol=Dataset-Unified Setting2026.04 | 67.32 | — | — | 67.74 | |
| EACLArchitecture Type=Sequence-based, Backbone=SimCSE-Roberta-Large2024.03 | 67.12 | — | — | — | |
| HidialogExtra Knowledge=✗, Model type=transformer, Method Setting=Small-scale Discriminant ERC-specific Model2023.09 | 66.96 | — | — | — | |
| CBERLBackbone=-, External FREE=-, Training Protocol=Dataset-Specific Setting2026.04 | 66.89 | — | — | 67.78 | |
| HyperEmo-LLaMA2-7BBackbone=LLaMA2-7B2026.05 | 66.8 | — | — | 67.9 | |
| MultiEMO (Original)reproduction=false2026.04 | 66.74 | — | — | — | |
| M2FNetModality=Multimodal2022.06 | 66.71 | — | — | 67.85 | |
| M2FNetYear=20222022.06 | 66.71 | — | — | — | |
| M2FNetExtra Knowledge=✗, Model type=transformer, Method Setting=Small-scale Discriminant ERC-specific Model2023.09 | 66.71 | — | — | — | |
| COMPMArchitecture Type=Sequence-based2024.03 | 66.52 | — | — | — | |
| MPLPArchitecture Type=Sequence-based2024.03 | 66.51 | — | — | — | |
| EmotionFlowYear=20222022.06 | 66.5 | — | — | — | |
| COMPM + SABase Model=COMPM, Grayscale Label Type=self-adjust2022.06 | 66.5 | — | — | — | |
| RoBERTa + FSABase Model=RoBERTa, Grayscale Label Type=future-self-adjust2022.06 | 66.49 | — | — | — | |
| SACLArchitecture Type=Sequence-based2024.03 | 66.45 | — | — | — | |
| SACL-LSTMParameters=2.6M, Extra Knowledge=✗, Model type=recurrent, Method Setting=Small-scale Discriminant ERC-specific Model2023.09 | 66.45 | — | — | — | |
| DIEUArchitecture Type=Sequence-based2024.03 | 66.43 | — | — | — | |
| EmotionICExtra Knowledge=✗, Model type=recurrent, Method Setting=Small-scale Discriminant ERC-specific Model2023.09 | 66.4 | — | — | — | |
| TODKAT (-Topics)Average of ten runs=true, latent topics=false2021.06 | 66.34 | — | 63.52 | — | |
| MultiEMO (Rep.)reproduction=true2026.04 | 66.34 | — | — | — | |
| HCANParameters=3.5M, Extra Knowledge=✗, Model type=recurrent, Method Setting=Small-scale Discriminant ERC-specific Model2023.09 | 66.24 | — | — | — | |
| COSMIC+CKCLArchitecture Type=Sequence-based2024.03 | 66.21 | — | — | — | |
| SIGATArchitecture Type=Graph-based2024.03 | 66.2 | — | — | — | |
| COMPM + SBase Model=COMPM, Grayscale Label Type=self2022.06 | 66.19 | — | — | — | |
| LlamaParameters=12.5M(7B), Extra Knowledge=✗, Model type=LLM-based, Method Setting=LoRA + Backbone2023.09 | 66.15 | — | — | — | |
| SPCLParameters=356.7M, Extra Knowledge=✗, Model type=transformer, Method Setting=Small-scale Discriminant ERC-specific Model2023.09 | 66.13 | — | — | — | |
| DER-GCNBackbone=-, External FREE=-, Training Protocol=Dataset-Specific Setting2026.04 | 66.1 | — | — | 66.8 | |
| RoBERTa + SABase Model=RoBERTa, Grayscale Label Type=self-adjust2022.06 | 66 | — | — | — | |
| Llama2Parameters=12.5M(7B), Extra Knowledge=✗, Model type=LLM-based, Method Setting=LoRA + Backbone2023.09 | 65.84 | — | — | — | |
| KATCOMETAverage of ten runs=true, knowledge base=COMET2021.06 | 65.82 | — | 63.07 | — | |
| HCAMModality=Audio+Text2023.04 | 65.8 | — | — | — | |
| DialogueCRNParameters=3.3M, Extra Knowledge=✗, Model type=recurrent, Method Setting=Small-scale Discriminant ERC-specific Model2023.09 | 65.77 | — | — | — | |
| SPCL+CLArchitecture Type=Sequence-based, Backbone=SimCSE-Roberta-Large2024.03 | 65.74 | — | — | — | |
| RoBERTa + SBase Model=RoBERTa, Grayscale Label Type=self2022.06 | 65.71 | — | — | — | |
| SCCLArchitecture Type=Sequence-based2024.03 | 65.7 | — | — | — | |
| SupConArchitecture Type=Sequence-based2024.03 | 65.63 | — | — | — | |
| EmoBERTaYear=20212022.06 | 65.61 | — | — | — | |
| EmoBERTaParameters=355M, Extra Knowledge=✗, Model type=transformer, Method Setting=Small-scale Pretrained Language Model2023.09 | 65.61 | — | — | — | |
| HCAMModality=Text2023.04 | 65.6 | — | — | — | |
| LR-GCNBackbone=-, External FREE=-, Training Protocol=Dataset-Specific Setting2026.04 | 65.6 | — | — | — | |
| ChatGLM2Parameters=12.5M(6B), Extra Knowledge=✗, Model type=LLM-based, Method Setting=LoRA + InstructERC2023.09 | 65.58 | — | — | — | |
| EGMFBackbone=GLM3-6B2026.01 | 65.57 | — | — | 67.22 | |
| COMPMBase Model=COMPM, Grayscale Label Type=None2022.06 | 65.53 | — | — | — | |
| UniMSEParameters=220M, Extra Knowledge=✗, Model type=transformer, Method Setting=Small-scale Pretrained Language Model2023.09 | 65.51 | — | — | — | |
| UniMSE2026.04 | 65.51 | — | — | — | |
| UniMSE2026.05 | 65.51 | — | — | 65.09 | |
| RoBERTa + WBase Model=RoBERTa, Grayscale Label Type=word-embedding2022.06 | 65.48 | — | — | — | |
| TODKATYear=20212022.06 | 65.47 | — | — | — | |
| TODKATParameters=330M, Extra Knowledge=COMET, Model type=transformer, Method Setting=Small-scale Discriminant ERC-specific Model2023.09 | 65.47 | — | — | — | |
| EGMFBackbone=llama2-7B2026.01 | 65.42 | — | — | 66.46 | |
| DialogueEINArchitecture Type=Sequence-based2024.03 | 65.37 | — | — | — | |
| DialogueRNNParameters=9.9M, Extra Knowledge=✗, Model type=recurrent, Method Setting=Small-scale Discriminant ERC-specific Model2023.09 | 65.3 | — | — | — | |
| COG-BART + SABase Model=COG-BART, Grayscale Label Type=self-adjust2022.06 | 65.28 | — | — | — | |
| RoBERTa + CBase Model=RoBERTa, Grayscale Label Type=category2022.06 | 65.27 | — | — | — | |
| Psychological + SBase Model=Psychological, Grayscale Label Type=self2022.06 | 65.24 | — | — | — | |
| COSMIC2021.06 | 65.21 | — | 65.28 | — | |
| COSMICYear=20202022.06 | 65.21 | — | — | — | |
| COSMICArchitecture Type=Sequence-based2024.03 | 65.21 | — | — | — | |
| EGMFBackbone=GLM4-9B2026.01 | 65.21 | — | — | 67.01 | |
| COSMICBackbone=-, External FREE=-, Training Protocol=Dataset-Specific Setting2026.04 | 65.21 | — | — | — | |
| SKAIGExtra Knowledge=COMET, Model type=GNN, Method Setting=Small-scale Discriminant ERC-specific Model2023.09 | 65.18 | — | — | — | |
| Psychological + SABase Model=Psychological, Grayscale Label Type=self-adjust2022.06 | 65.04 | — | — | — | |
| EGMFBackbone=llama3-8B2026.01 | 65.04 | — | — | 66.42 | |
| MGLRA2026.01 | 64.9 | — | — | 66.4 | |
| CauAINParameters=6.1M, Extra Knowledge=ATOMIC, Model type=recurrent, Method Setting=Small-scale Discriminant ERC-specific Model2023.09 | 64.89 | — | — | — | |
| ChatGLM2Parameters=12.5M(6B), Extra Knowledge=✗, Model type=LLM-based, Method Setting=LoRA + Backbone2023.09 | 64.85 | — | — | — | |
| CoG-BART2021.12 | 64.81 | — | 65.95 | — | |
| Cog-BARTArchitecture Type=Sequence-based2024.03 | 64.81 | — | — | — | |
| RoBERTa (ours)Base Model=RoBERTa, Grayscale Label Type=None2022.06 | 64.79 | — | — | — | |
| COG-BARTBase Model=COG-BART, Grayscale Label Type=None2022.06 | 64.77 | — | — | — | |
| KATSBERTAverage of ten runs=true, knowledge base=SBERT2021.06 | 64.54 | — | 61.88 | — | |
| Lian et al.Modality=Audio+Text2023.04 | 64.5 | — | — | — | |
| PsychologicalBase Model=Psychological, Grayscale Label Type=None2022.06 | 64.38 | — | — | — | |
| COG-BART + SBase Model=COG-BART, Grayscale Label Type=self2022.06 | 64.26 | — | — | — | |
| LSDGNN2026.01 | 64.07 | — | — | 64.67 | |
| LSDGNN+ICLBackbone=-, External FREE=-, Training Protocol=Dataset-Specific Setting2026.04 | 64.07 | — | — | 64.67 | |
| Xie et al.Modality=Multimodal2022.06 | 64 | — | — | 65 | |
| Li et al.Modality=Audio+Text2023.04 | 64 | — | — | — | |
| EmoCaps2026.01 | 64 | — | — | — | |
| MultiDAG+CLBackbone=-, External FREE=-, Training Protocol=Dataset-Specific Setting2026.04 | 64 | — | — | — | |
| EmoCaps2026.05 | 64 | — | — | — | |
| TODKAT (-KB)Average of ten runs=true, knowledge base=false2021.06 | 63.97 | — | 61.11 | — | |
| DAG-ERC+HCLArchitecture Type=Graph-based2024.03 | 63.89 | — | — | — | |
| PRC-EmoBackbone=Qwen3-8B, External FREE=NO, Training Protocol=Dataset-Unified Setting2026.04 | 63.88 | — | — | 64.07 | |
| CoG-BARTParameters=415.1M, Extra Knowledge=✗, Model type=transformer, Method Setting=Small-scale Discriminant ERC-specific Model2023.09 | 63.82 | — | — | — | |
| DAG-ERCYear=20212022.06 | 63.65 | — | — | — | |
| DAG-ERCArchitecture Type=Graph-based2024.03 | 63.65 | — | — | — | |
| DAG-ERCBase Model=DAG-ERC, Grayscale Label Type=None2022.06 | 63.63 | — | — | — | |
| DAG-ERCBackbone=-, External FREE=-, Training Protocol=Dataset-Specific Setting2026.04 | 63.63 | — | — | 63.98 | |
| Mao et al.Modality=Audio+Text2023.04 | 63.6 | — | — | — | |
| BART-largeBackbone=BART-large2021.12 | 63.57 | — | 64.41 | — |