Automated Hate Speech Detection and the Problem of Offensive Language

TL;DR

This paper addresses the challenge of accurately distinguishing hate speech from offensive language on social media by developing a multi-class classifier trained on crowd-sourced labeled data, revealing specific difficulties in classification.

Contribution

It introduces a novel multi-class classification approach using crowd-sourced labels to differentiate hate speech from offensive language, improving upon lexical and previous supervised methods.

Findings

Racist and homophobic tweets are more reliably classified as hate speech.

Sexist tweets tend to be classified as offensive rather than hate speech.

Tweets without explicit hate keywords are more challenging to classify.

Abstract

A key challenge for automatic hate-speech detection on social media is the separation of hate speech from other instances of offensive language. Lexical detection methods tend to have low precision because they classify all messages containing particular terms as hate speech and previous work using supervised learning has failed to distinguish between the two categories. We used a crowd-sourced hate speech lexicon to collect tweets containing hate speech keywords. We use crowd-sourcing to label a sample of these tweets into three categories: those containing hate speech, only offensive language, and those with neither. We train a multi-class classifier to distinguish between these different categories. Close analysis of the predictions and the errors shows when we can reliably separate hate speech from other offensive language and when this differentiation is more difficult. We find that racist and homophobic tweets are more likely to be classified as hate speech but that sexist tweets are generally classified as offensive. Tweets without explicit hate keywords are also more difficult to classify.