dataset_info:
features:
- name: sentence1
dtype: string
- name: sentence2
dtype: string
- name: labels
list: float64
- name: metadata
dtype: string
splits:
- name: train
num_bytes: 1763729
num_examples: 1282
- name: test
num_bytes: 1800106
num_examples: 1288
download_size: 1565152
dataset_size: 3563835
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
license: other
source_datasets: []
discogem
DiscoGeM 1.0 native crowd vote shares and original train/dev/test labels.
Recovers integer votes from rounded majority_softlabel and observations; validates their sum. No sampled tie-breaking, IRT or model probabilities. Keeps context-containing argument spans and actual observation counts. Rows marked exclude are omitted. Shared source paragraphs remain in their native highest-priority holdout (test, dev, train); this empties the overlapping dev partition rather than fabricating one. Original distribution: https://github.com/merelscholman/DiscoGeM. Underlying source texts have mixed terms; overall dataset license is unspecified.
Original data: original distribution. Repackaged as parquet for tasksource by scripts/repackage_dataset/stance.py.
Conversion provenance: source revisions, hashes, configuration and exclusions.