Giters

shenwzh3 / DAG-ERC

Pytorch code for ACL-IJCNLP accepted paper "Directed Acyclic Graph Network for Conversational Emotion Recognition"

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

The label value

Tars-is-a-robot opened this issue · comments

commented

I'm sorry to ask you，Why do some non-populated data in IEMOCAP data sets have a label value of -1

commented

That's because IEMOCAP only uses 6 emotions for the evaluation('hap', 'sad', 'neu', 'ang', 'exc', 'fru'), which means they discard those labels which rarely appear in the IEMOCAP dataset by setteing them to -1(disgust, fear, other, surprise).
But the dataset seems to have less effective labels than it describes in the article, since the article says it has 5810 utterances for train and validation and 1623 for test, but actually, there are only 5758(4778 for train and 980 for valid) for train and validation and 1622 for test.