dataset_info:
features:
- name: sentence1
dtype: string
- name: sentence2
dtype: string
- name: labels
dtype: string
- name: metadata
dtype: string
splits:
- name: train
num_bytes: 3894325
num_examples: 9911
- name: test
num_bytes: 1052776
num_examples: 2675
download_size: 1115339
dataset_size: 4947101
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
license: cc-by-4.0
source_datasets: []
uk-election
Human target-specific sentiment from the EACL 2017 UK-election corpus.
Joins original tweet text, final annotations and official train/test IDs. Released split IDs omit a constant 500000000000000000; a complete bijection to all 4,077 tweet IDs is required before joining. Entity/keyword targets and human-added targets are retained; overall topic judgments, doesnotapply and invalid character spans are excluded. No synthetic labels or holdouts. Annotations: CC BY 4.0; original tweet content remains under Twitter's terms. Original distribution: https://doi.org/10.6084/m9.figshare.4479563.v1.
Original data: original distribution. Repackaged as parquet for tasksource by scripts/repackage_dataset/target_sentiment.py.
Conversion provenance: source revisions, hashes, configuration and exclusions.