Country

United Kingdom United Kingdom

Title

Vidgen, B. and Yasseri, T. (2018), Detecting weak and strong Islamophobic hate speech on social media

View full Research

Year

2018

Type of publication

Report

Geographical coverage

National

Area/location of interest

Not applicable - national level

Type of Institution

Academic

Institution

UK, University of Oxford

Main Thematic Focus

Hate crime Incitement to hatred or violence

Target Population

Muslims Religious minorities

Key findings

Having acknowledged the harm that Islamophobic hate speech on social media causes to individuals and society as a whole, this research aims to development a tool to detect and classify Islamophobic hate speech.

The multi-class classifier that was developed distinguishes between non-Islamophobic, weak Islamophobic and strong Islamophobic content with 77.6% accuracy and 83% balanced accuracy.

Methodology (Qualitative/Quantitative and exact type used, questionnaires etc)

Quantative

Sample details and representativeness

The dataset consisted of 140 million tweets posted in 2017 and the first six months of 2018 by followers of mainstream and far0right political parties.

• 7,500 users randomly selected from followers of UKIP.
• 7,500 users randomly selected from followers of the Conservatives.
• 7,500 users randomly selected from followers of Labour.
• 7,500 users randomly selected from followers of the Liberal Democrats.
• All ~15,000 followers of the BNP.
• All 32,000 followers of Britain First.
• Every tweet produced by a set of 45 far right accounts (consisting of every group which appear in Hope Not Hate’s 2015 and 2017 reports on the far right.
• Every @ mention of the same 45 far right accounts.

DISCLAIMERThe information presented here is collected under contract by the FRA's research network FRANET. The information and views contained do not necessarily reflect the views or the official position of the FRA.