README.md

metadata

configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
  - config_name: embedding_all-MiniLM-L12-v2
    data_files:
      - split: train
        path: embedding_all-MiniLM-L12-v2/train-*
      - split: validation
        path: embedding_all-MiniLM-L12-v2/validation-*
      - split: test
        path: embedding_all-MiniLM-L12-v2/test-*
  - config_name: embedding_all-mpnet-base-v2
    data_files:
      - split: train
        path: embedding_all-mpnet-base-v2/train-*
      - split: validation
        path: embedding_all-mpnet-base-v2/validation-*
      - split: test
        path: embedding_all-mpnet-base-v2/test-*
  - config_name: embedding_multi-qa-mpnet-base-dot-v1
    data_files:
      - split: train
        path: embedding_multi-qa-mpnet-base-dot-v1/train-*
      - split: validation
        path: embedding_multi-qa-mpnet-base-dot-v1/validation-*
      - split: test
        path: embedding_multi-qa-mpnet-base-dot-v1/test-*
dataset_info:
  - config_name: default
    features:
      - name: id
        dtype: string
      - name: text
        dtype: string
      - name: labels
        dtype:
          class_label:
            names:
              '0': non
              '1': tox
      - name: uid
        dtype: int64
    splits:
      - name: train
        num_bytes: 55430581
        num_examples: 127656
      - name: validation
        num_bytes: 13936861
        num_examples: 31915
      - name: test
        num_bytes: 27474227
        num_examples: 63978
    download_size: 62548640
    dataset_size: 96841669
  - config_name: embedding_all-MiniLM-L12-v2
    features:
      - name: uid
        dtype: int64
      - name: embedding_all-MiniLM-L12-v2
        sequence: float32
    splits:
      - name: train
        num_bytes: 197611488
        num_examples: 127656
      - name: validation
        num_bytes: 49404420
        num_examples: 31915
      - name: test
        num_bytes: 99037944
        num_examples: 63978
    download_size: 484421377
    dataset_size: 346053852
  - config_name: embedding_all-mpnet-base-v2
    features:
      - name: uid
        dtype: int64
      - name: embedding_all-mpnet-base-v2
        sequence: float32
    splits:
      - name: train
        num_bytes: 393691104
        num_examples: 127656
      - name: validation
        num_bytes: 98425860
        num_examples: 31915
      - name: test
        num_bytes: 197308152
        num_examples: 63978
    download_size: 827919212
    dataset_size: 689425116
  - config_name: embedding_multi-qa-mpnet-base-dot-v1
    features:
      - name: uid
        dtype: int64
      - name: embedding_multi-qa-mpnet-base-dot-v1
        sequence: float32
    splits:
      - name: train
        num_bytes: 393691104
        num_examples: 127656
      - name: validation
        num_bytes: 98425860
        num_examples: 31915
      - name: test
        num_bytes: 197308152
        num_examples: 63978
    download_size: 827907964
    dataset_size: 689425116
annotations_creators:
  - crowdsourced
language:
  - en
language_creators:
  - found
license:
  - cc0-1.0
multilinguality:
  - monolingual
pretty_name: Toxic Wikipedia Comments
size_categories:
  - 100K<n<1M
source_datasets:
  - extended|other
tags:
  - wikipedia
  - toxicity
  - toxic comments
task_categories:
  - text-classification
task_ids:
  - hate-speech-detection

This dataset has been created as an artefact of the paper AnchorAL: Computationally Efficient Active Learning for Large and Imbalanced Datasets (Lesci and Vlachos, 2024). More info about this dataset in the appendix of the paper. This is the same dataset as OxAISH-AL-LLM/wiki_toxic. The only differences are:

1. Addition of a unique identifier, uid.

2. Addition of the indices, that is, 3 columns with the embeddings of 3 different sentence-transformers

   • all-mpnet-base-v2
   • multi-qa-mpnet-base-dot-v1
   • all-MiniLM-L12-v2

3. Renaming of the label column to labels for easier compatibility with the transformers library.