Skip to main content

How to Perform Hate Speech Analysis using DataLab

DataLab supports different types of bias analyses for datasets and hate speech is one case. Specifically, given a dataset, DataLab can identify what percentage of samples contains hate speech words. Although deciding whether a sentence contains toxic language is a slightly complex task, which may involve the confounding effects of dialect and the social identity of a speaker Sap et al. (2019), we make a first step by following Davidson et al (2017), classifying the samples into following categories:

  • hate speech
  • offensive language
  • neither

We then calculate the ratio of samples in different categories.

To perform this type of analysis:

1. Dataset Selection​

You just need to choose a dataset and click the right mouse button, and choose analysis -> bias, then you will enter into a page designed for bias analysis

2. Choose the hate speech filter​

As shown below, different colors represent the proportions of samples with different categories (hate speech, offensive language and neither)