license: mit
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: validation
path: data/validation-*
- split: test
path: data/test-*
dataset_info:
features:
- name: premise
dtype: string
- name: target
dtype: string
- name: stance
dtype: int64
- name: hypothesis
dtype: string
- name: entailment
dtype: int64
splits:
- name: train
num_bytes: 869570
num_examples: 1935
- name: validation
num_bytes: 288178
num_examples: 645
- name: test
num_bytes: 293856
num_examples: 645
download_size: 464876
dataset_size: 1451604
This dataset contains quote tweets that have been hand labeled for stance towards a politician. Quote tweets are a particularly challenging classification task because they contain multiple (often contradictory) expressions from multiple authors.
Twitter handles from politicians in the dataset have been replaced by their name. Be aware that "rt @realdonaldtrump You're a liar!" has been replaced with "rt trump You're a liar!", And means the author is retweeting Trump calling someone else a liar, not the author calling Trump a liar.
Stance:
-1: Against: The document is critical of the target.
0: Neutral: The document doesn't express an opinion about the target or it can't be determined what the expressed opinion is with the given context.
1: Support: The document expresses support for the target. Expressing collaboration on bills or letters is considered support.
Label:
0: Entail
1: Not Entail
The test set contains 645 documents, ~400 of which are were randomly sampled from the entire data set and the rest are about 6 politicians not included in the training or validation data. These politicians are:
- Ted Cruz
- Hakeem Jeffries
- Madison Cawthorn
- Alexandria Ocasio-Cortez
- Mitt Romney
- Kyrsten Sinema