Tudorx95/NER_Political_Economic

NER Political & Economic Dataset

Custom NER dataset for politico-economic entity recognition, built using:

• Pre-annotated sources: CoNLL-2003, WNUT-2017 (re-mapped via gazetteers)
• Weak supervision: Snorkel labeling functions over CC-News, Wikipedia, SEC EDGAR
• Synthetic templates for rare classes

Schema (11 labels)

POLITICIAN, POLITICAL_PARTY, POLITICAL_ORG, FINANCIAL_ORG, ECONOMIC_INDICATOR, POLICY, LEGISLATION, MARKET_EVENT, CURRENCY, TRADE_AGREEMENT, GPE

Splits

• splits/train.jsonl — training set
• splits/dev.jsonl — validation set
• splits/test.jsonl — test set (includes CoNLL-2003 test as gold standard)

Format

Each line is a JSON object:

{
    "text": "The Federal Reserve raised rates.",
    "entities": [
        {"text": "Federal Reserve", "label": "FINANCIAL_ORG", "start": 4, "end": 19}
    ],
    "source": "weak_supervision_cc_news"
}

Methodology

Inspired by Lison et al. (2020), "Named Entity Recognition without Labelled Data: A Weak Supervision Approach" (ACL 2020).