Anti-Islamic Content on Twitter

This blogpost was written by Carl Miller, Research Director, and Josh Smith, Researcher, at the Centre for the Analysis of Social Media (CASM) at Demos. @carljackmiller

This analysis was presented at the Mayor of London’s Policing and Crime Summit on Monday 24 April, 2017.

The Centre for the Analysis of Social Media at Demos has been conducting research to measure the volume of messages on Twitter algorithmically considered to be derogatory towards Muslims over a year, from March 2016 to March 2017. This is part of a broad effort to understand the scale, scope and nature of uses of social media that are possibly socially problematic and damaging.

Over a year, Demos’ researchers detected 143,920 Tweets sent from the UK considered to be derogatory and anti-Islamic – this is about 393 a day. These Tweets were sent from over 47,000 different users, and fell into a number of different categories – from directed insults to broader political statements.

A random sample of hateful Tweets were manually classified into three broad categories:

• ‘Insult’ (just under half): Tweets used an anti-Islamic slur in a derogatory way, often directed at a specific individual.
• ‘Muslims are terrorists’(around one fifth) Derogatory statements generally associating Muslims and Islam with terrorism.
• ‘Muslims are the enemy’ (just under two fifths): Statements claiming that Muslims, generally, are dedicated toward the cultural and social destruction of the West.

The researchers found that key events, especially terrorist attacks, drive large increases in the volume of messages on Twitter containing this kind of language.

The Brussels, Orlando, Nice, Normandy, Berlin and Quebec attacks all caused large increases. There was a period of heightened activity over Brexit, and sometimes online ‘Twitter storms’ (such as the use of derogatory slurs by Azealia Banks toward Zayn Malik) also drove sharp increases.

Tweets containing this language were sent from every region of the UK, but the most over-represented areas, compared to general Twitter activity, were London and the North-West.

Of the 143,920 Tweets containing this language and classified as being sent from within the UK, 69,674 (48%) contained sufficient information to be located within a broad area of the UK. To measure how many Tweets each region generally sends, a random baseline of 67 Million Tweets were collected over 19 days over late February and early March. The volume of Tweets containing derogatory language towards Muslims was compared to this baseline. This identified regions where the volume was higher or lower than the expectation on the basis of general activity on Twitter.

In London, North London sent markedly more tweets containing language considered derogatory towards Muslims than South London.

27,576 (39%) tweets were sent from Greater London. Of these, 14,953 Tweets (about half) could be located to a more specific region within London (called a ‘NUTS-3 region’; typically either a London Borough or a combination of a small number of London Boroughs).^[1]

• Brent, Redbridge and Waltham Forest sent the highest number of derogatory, anti-Islamic Tweets relative to their baseline average of general Twitter activity.
• Westminster and Bromley sent the least number of derogatory, anti-Islamic Tweets relative to their baseline average of general Twitter activity.

Demos’ research identified six different online tribes. ^[2] These were:

Core political anti-Islam. The largest group of about 64,000 users including recipients of Tweets. Politically active group engaged with international politics.

• Hashtags employed by this group suggest engagement in anti-Islam and right wing political conversations: (#maga #tcot #auspol #banIslam #stopIslam #rapefugees)
• In aggregate, words in user descriptions emphasise nationality, right-wing political interest and hostility towards Islam (anti, Islam, Brexit, UKIP, proud, country)

Contested reactions to Terrorist attacks. The second largest group, of about 18,000 users, including recipients of tweets.

• Aggregate overview of user descriptions imply a relatively young group (sc, snapchat, ig, instagram, 17,18,19,20, 21)
• User descriptions also imply a mix of political opinion (blacklivesmatter, whitelivesmatter, freepalestine)
• Hashtags engage in conversations emerging in the aftermath of terrorist attacks (#prayforlondon, #munich, #prayforitaly, #prayforistabul, #prayformadinah, #orlando)
• Likewise, hashtags are a mix of pro- and anti-Islamic (#britainfirst, #whitelivesmatter, #stopislam, #postrefracism, #humanity)

The counter-speechers. A group of 8,700 people; although of course the data collection, by design, only detected the part of the counter-speech conversation containing language that can be used in a way derogatory towards Muslims. It is therefore likely that it did not collect the majority of counter-speech activity.^[3]

The shape of the cluster shows a smaller number of highly responded to-/retweeted comments.

• Hashtags engage predominantly with anti-racist conversations (#racisttrump, postrefracism, #refugeeswelcome, #racism, #islamophobia)
• In aggregate, user descriptions show mix of political engagement and general identification with left-wing politics (politics, feminist, socialist, Labour).
• Overall they also show more descriptions of employment than the other clusters (writer, author, journalist, artist).

The Football Fans. 7,530 users are in this cluster, including recipients of Tweets.

• The bio descriptions of users within his cluster overwhelmingly contain football-related words (fan, football, fc, lfc, united, liverpool, arsenal, support, club, manchester, mufc, chelsea, manutd, westham)
• No coherent use of hashtags. This cluster engaged in lots of different conversations.

India/Pakistan. Just under 5,000 users are in this cluster (including recipients).

• Hashtags overwhelmingly engage in conversation to do with India-Pakistan relations or just Pakistan (#kashmir, #surgicalstrike, #pakistan, #actagainstpak).
• In aggregate, words in user descriptions relate to Indian/nationalist identity and pro-Modi identification (proud, Indian, hindu, proud indian, nationalist, dharma, proud hindu, bhakt,)

The Gamers. 2,813 users are in this cluster (including Tweet recipients).

• There is no coherent use of hashtags.
• Overall, aggregate comments in user descriptions either imply young age (16,17,18) or are related to gaming (player, cod [for ‘Call of Duty’], psn)

A small number of accounts overall are responsible for many of the tweets containing language generally considered to be derogatory towards Muslims.

• 50% of Tweets classified as containing language considered anti-Islamic and derogatory are sent by only 6% of accounts
• 25% of Tweets classified as containing language considered anti-Islamic and derogatory were sent by 1% of accounts.

Likewise, a small number of accounts were the recipients of the derogatory, anti-Islamic activity that was directed at a particular person.

The full paper, outlining methodology and ethical notes, can be downloaded here. 



NOTES –
^[1] An important caveat is that the volumes associated with each of these regions are obviously smaller than the total number of Tweets in the dataset overall
^[2] A caveat here is that this network graph includes Tweets that are misclassified and also includes the recipients of abuse. It is also important to note that not everyone who shares Tweets does so with malicious intent; they can be doing so to highlight the abuse to their own followers.
^[3] In other work on the subject we have found there are usually more posts about solidarity, support for Muslims than attacks on them.