metadata
dataset_info:
features:
- name: sentence1
dtype: string
- name: sentence2
dtype: string
- name: labels
dtype: string
- name: metadata
dtype: string
splits:
- name: train
num_bytes: 121756713
num_examples: 49753
- name: test
num_bytes: 58697230
num_examples: 25413
download_size: 102001021
dataset_size: 180453943
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
license: other
source_datasets: []
fnc1
FNC-1 native headline/body stance, joined from the original labeled CSVs.
Preserves complete article bodies and official training/competition-test partitions. Does not invent validation data. Identical body copies remain in the native holdout. Original distribution: https://github.com/FakeNewsChallenge/fnc-1. The original distribution does not specify a dataset license.
Original data: original distribution. Repackaged as parquet for tasksource by scripts/repackage_dataset/stance.py.
Conversion provenance: source revisions, hashes, configuration and exclusions.