metadata

language:
  - de
license: cc-by-4.0
task_categories:
  - text-classification
pretty_name: GermEval 18
dataset_info:
  features:
    - name: text
      dtype: string
    - name: coarse
      dtype:
        class_label:
          names:
            '0': OTHER
            '1': OFFENSE
    - name: fine
      dtype:
        class_label:
          names:
            '0': OTHER
            '1': ABUSE
            '2': INSULT
            '3': PROFANITY
  splits:
    - name: train
      num_bytes: 826320
      num_examples: 5009
    - name: test
      num_bytes: 509105
      num_examples: 3532
  download_size: 867329
  dataset_size: 1335425
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*

GermEval18 Loader

Data Repository: https://github.com/uds-lsv/GermEval-2018-Data
Data Reference: https://doi.org/10.11588/data/0B5VML
Paper: https://epub.oeaw.ac.at/0xc1aa5576_0x003a10d2.pdf

Info

Note: This dataset is a loader script that pulls the data straight from the official GitHub repository.

What is the difference to philschmid/germeval18?: We did not get all samples, when using the former script.

Output from philschmid/germeval18:

DatasetDict({
  train: Dataset({
      features: ['text', 'binary', 'multi'],
      num_rows: 5009
  })
  test: Dataset({
      features: ['text', 'binary', 'multi'],
      num_rows: 3398
  })
})

but the dataset (that our loader script is based on) contains all samples from the GermEval18 dataset:

DatasetDict({
  train: Dataset({
      features: ['text', 'coarse', 'fine'],
      num_rows: 5009
  })
  test: Dataset({
      features: ['text', 'coarse', 'fine'],
      num_rows: 3532
  })
})