Hugging Face's logo

Datasets:
imvladikon
/
english_news_weak_ner

Tasks:
Token Classification
Modalities:
Text
Time-series
Formats:
parquet
Languages:
English
Size:
1M - 10M
Libraries:
Datasets
Dask
Dataset card Viewer Files Community
2
english_news_weak_ner / README.md
imvladikon's picture
imvladikon
Upload README.md with huggingface_hub
9fbfc78
preview code
|
raw
history blame
39.5 kB
metadata
language:
  - en
size_categories:
  - 1M<n<10M
task_categories:
  - token-classification
dataset_info:
  - config_name: articles
    features:
      - name: title
        dtype: string
      - name: author
        dtype: string
      - name: datetime
        dtype: string
      - name: url
        dtype: string
      - name: month
        dtype: string
      - name: day
        dtype: string
      - name: doc_id
        dtype: string
      - name: text
        dtype: string
      - name: year
        dtype: string
      - name: doc_title
        dtype: string
    splits:
      - name: train
        num_bytes: 1313871812
        num_examples: 446809
    download_size: 791316510
    dataset_size: 1313871812
  - config_name: entities
    features:
      - name: doc_id
        dtype: string
      - name: sent_num
        dtype: int32
      - name: sentence
        dtype: string
      - name: doc_title
        dtype: string
      - name: spans
        sequence:
          - name: Score
            dtype: float32
          - name: Type
            dtype: string
          - name: Text
            dtype: string
          - name: BeginOffset
            dtype: int32
          - name: EndOffset
            dtype: int32
      - name: tags
        struct:
          - name: tokens
            sequence: string
          - name: raw_tags
            sequence: string
          - name: ner_tags
            sequence:
              class_label:
                names:
                  '0': B-DATE
                  '1': I-DATE
                  '2': L-DATE
                  '3': U-DATE
                  '4': B-DUC
                  '5': I-DUC
                  '6': L-DUC
                  '7': U-DUC
                  '8': B-EVE
                  '9': I-EVE
                  '10': L-EVE
                  '11': U-EVE
                  '12': B-LOC
                  '13': I-LOC
                  '14': L-LOC
                  '15': U-LOC
                  '16': B-MISC
                  '17': I-MISC
                  '18': L-MISC
                  '19': U-MISC
                  '20': B-ORG
                  '21': I-ORG
                  '22': L-ORG
                  '23': U-ORG
                  '24': B-PER
                  '25': I-PER
                  '26': L-PER
                  '27': U-PER
                  '28': B-QTY
                  '29': I-QTY
                  '30': L-QTY
                  '31': U-QTY
                  '32': B-TTL
                  '33': I-TTL
                  '34': L-TTL
                  '35': U-TTL
                  '36': O
    splits:
      - name: train
        num_bytes: 3665237140
        num_examples: 3515149
    download_size: 967133582
    dataset_size: 3665237140
configs:
  - config_name: articles
    data_files:
      - split: train
        path: articles/train-*
  - config_name: entities
    data_files:
      - split: train
        path: entities/train-*

Large Weak Labelled NER corpus

Dataset Summary

The dataset is generated through weak labelling of the scraped and preprocessed news corpus (bloomberg's news). so, only to research purpose. In order of the tokenization, news were splitted into sentences using nltk.PunktSentenceTokenizer (so, sometimes, tokenization might be not perfect)

Usage 

from datasets import load_dataset

articles_ds = load_dataset("imvladikon/english_news_weak_ner", "articles") # just articles with metadata
entities_ds = load_dataset("imvladikon/english_news_weak_ner", "entities")

NER tags

Tags description:

  • O Outside of a named entity
  • PER Person
  • LOC Location
  • ORG Organization
  • MISC Miscellaneous
  • DATE Date and time expression
  • QTY Quantity
  • EVE Event
  • TTL Title
  • DUC Commercial item

Tags: 

['B-DATE', 'I-DATE', 'L-DATE', 'U-DATE', 'B-DUC', 'I-DUC', 'L-DUC', 'U-DUC', 'B-EVE', 'I-EVE', 'L-EVE', 'U-EVE', 'B-LOC', 'I-LOC', 'L-LOC', 'U-LOC', 'B-MISC', 'I-MISC', 'L-MISC', 'U-MISC', 'B-ORG', 'I-ORG', 'L-ORG', 'U-ORG', 'B-PER', 'I-PER', 'L-PER', 'U-PER', 'B-QTY', 'I-QTY', 'L-QTY', 'U-QTY', 'B-TTL', 'I-TTL', 'L-TTL', 'U-TTL', 'O']

Tags statistics: 

{
    "O": 281586813,
    "B-QTY": 2675754,
    "L-QTY": 2675754,
    "I-QTY": 2076724,
    "U-ORG": 1459628,
    "I-ORG": 1407875,
    "B-ORG": 1318711,
    "L-ORG": 1318711,
    "B-PER": 1254037,
    "L-PER": 1254037,
    "U-MISC": 1195204,
    "U-LOC": 1084052,
    "U-DATE": 1010118,
    "B-DATE": 919815,
    "L-DATE": 919815,
    "I-DATE": 650064,
    "U-PER": 607212,
    "U-QTY": 559523,
    "B-LOC": 425431,
    "L-LOC": 425431,
    "I-PER": 262887,
    "I-LOC": 201532,
    "I-MISC": 190576,
    "B-MISC": 162978,
    "L-MISC": 162978,
    "I-TTL": 64641,
    "B-TTL": 53330,
    "L-TTL": 53330,
    "B-EVE": 43329,
    "L-EVE": 43329,
    "U-TTL": 41568,
    "I-EVE": 35316,
    "U-DUC": 33457,
    "U-EVE": 19103,
    "I-DUC": 15622,
    "B-DUC": 15580,
    "L-DUC": 15580
}

Sample:

display

Articles:

{'title': 'Watson Reports Positive Findings for Prostate Drug',
 'author': 'RobertSimison',
 'datetime': '2007-01-16T14:16:56Z',
 'url': 'http://www.bloomberg.com/news/2007-01-16/watson-reports-positive-findings-for-prostate-drug-update1-.html',
 'month': '1',
 'day': '16',
 'doc_id': 'a5c7c556bd112ac22874492c4cdb18eb46a30905',
 'text': 'Watson Pharmaceuticals Inc. (WPI) , the\nlargest U.S. maker of generic drugs, reported positive results\nfor its experimental prostate treatment in two late-state trials.  \n The drug, silodosin, was more effective than a placebo in\ntreating enlarged prostates, or benign prostatic hyperplasia, the\nCorona, California-based company said today in a statement on PR\nNewswire. The tests were in the final of three phases of trials\nnormally needed for regulatory approval.  \n Non-cancerous enlarged prostate affects more than half of\nAmerican men in their 60s and as many as 90 percent of them by\nage 85, Watson said. Prescription drug sales to treat the\ndisorder total $1.7 billion a year, the company said.  \n Watson plans to apply for U.S. approval to market the drug\nin the first half of 2008, after completion later this year of a\none-year safety trial, the company said. The two studies reported\ntoday showed that cardiovascular and blood-pressure side effects\nwere low, Watson said.  \n To contact the reporter on this story:\nRobert Simison in  Washington  at \n rsimison@bloomberg.net .  \n To contact the editor responsible for this story:\nRobert Simison at   rsimison@bloomberg.net .',
 'year': '2007',
 'doc_title': 'watson-reports-positive-findings-for-prostate-drug-update1-'}

Entities:

{'doc_id': '806fe637ed51e03d9ef7a8889fc84f63f8fc8569',
 'sent_num': 9,
 'sentence': 'Spain and Portugal together accounted for 45\npercent of group profit in 2010.',
 'doc_title': 'bbva-may-post-lower-first-quarter-profit-hurt-by-spain-decline',
 'spans': {'Score': [0.7858654856681824,
   0.7856822609901428,
   0.9990736246109009,
   0.999079704284668],
  'Type': ['ORGANIZATION', 'ORGANIZATION', 'QUANTITY', 'DATE'],
  'Text': ['Spain', 'Portugal', '45\npercent', '2010'],
  'BeginOffset': [0, 10, 42, 72],
  'EndOffset': [5, 18, 52, 76]},
 'tags': {'tokens': ['Spain',
   'Spain',
   'and',
   'Portugal',
   'Spain',
   'and',
   'Portugal',
   'together',
   'accounted',
   'for',
   '45',
   '\n',
   'percent',
   'Spain',
   'and',
   'Portugal',
   'together',
   'accounted',
   'for',
   '45',
   '\n',
   'percent',
   'of',
   'group',
   'profit',
   'in',
   '2010',
   '.'],
  'raw_tags': ['U-ORG',
   'O',
   'O',
   'U-ORG',
   'O',
   'O',
   'O',
   'O',
   'O',
   'O',
   'B-QTY',
   'I-QTY',
   'L-QTY',
   'O',
   'O',
   'O',
   'O',
   'O',
   'O',
   'O',
   'O',
   'O',
   'O',
   'O',
   'O',
   'O',
   'U-DATE',
   'O'],
  'ner_tags': [23,
   36,
   36,
   23,
   36,
   36,
   36,
   36,
   36,
   36,
   28,
   29,
   30,
   36,
   36,
   36,
   36,
   36,
   36,
   36,
   36,
   36,
   36,
   36,
   36,
   36,
   3,
   36]}}

Data splits

name train
entities 3515149
articles 446809

Citation Information 

@misc{imvladikon2023bb_news_weak_ner,
  author = {Gurevich, Vladimir},
  title = {Weakly Labelled Large English NER corpus},
  year = {2022},
  howpublished = \url{https://huggingface.co/datasets/imvladikon/bloomberg_news_weak_ner},
}