EmoWOZ: A Large-Scale Corpus and Labelling Scheme for Emotion Recognition in Task-Oriented Dialogue Systems

Benchmark Model Rank Results
emotion-recognition-in-conversation-on-emowozCOSMIC#1Macro F1 (w/o Neutral): 56.34Weighted F1 (w/o Neutral): 77.09
emotion-recognition-in-conversation-on-emowozContextBERT#2Macro F1 (w/o Neutral): 54.30Weighted F1 (w/o Neutral): 79.67
emotion-recognition-in-conversation-on-emowozDialogueRNN-BERT#3Macro F1 (w/o Neutral): 52.15Weighted F1 (w/o Neutral): 75.50
emotion-recognition-in-conversation-on-emowozBERT#4Macro F1 (w/o Neutral): 50.14Weighted F1 (w/o Neutral): 73.55
emotion-recognition-in-conversation-on-emowozDialogueRNN-GloVe#5Macro F1 (w/o Neutral): 40.14Weighted F1 (w/o Neutral): 74.56