Skip to content

KLUE STS dataset description

Jihyung Moon edited this page May 27, 2021 · 1 revision
Name Description
key value
guid unique identifier
source document source of the sentences; 'source corpus-pairing method'
sentence1 first sentence input
sentence2 second sentence input
labels three types of labels for training and evaluation
- round1 real number label rounded up to the first decimal digit
- real_label original real number label
- binary_label binarized label with the basis of 3
annotations metadata for annotation
- agreement the distribution of annotations from 0 to 5
- annotators annotator ids of the sentence pair
- annotations annotation per annotator id