Actions

WikiHarass

From Algolit

Type: Dataset
Number of words: 1.039.789
Unique words: 64.136
Source: English Wikipedia
Developed by: Wikimedia Foundation

The Detox dataset is a project by Wikimedia and Perspective API to train a neural network that would detect the level of toxicity of a comment.

The original dataset consists of:

  • A corpus of all 95 million user and article talk diffs made between 2001 and 2015 scored by the personal attack model.
  • A human annotated dataset of 1m crowd-sourced annotations that cover 100k talk page diffs (with 10 judgements per diff).

For Algolit, a smaller section of the Detox dataset was used, taken from Jigsaw's Github, which contains both constructive and vandalist edits.