How Twitter Network Features Predict Users' Attitudes towards Islam

Next up at Web Science 2016 is Walid Magdy, whose focus is on social media commentary following the terrorist attacks in Paris in late 2016. Immediately after the attacks, sympathy with Paris was expressed on Twitter – but as the attacks were linked with Islamist terrorists, anti-Muslim messages also began to appear.

Walid's team tracked relevant keywords and hashtags on Twitter after the attacks, and noted that more than one tenth of all messages discussed Islam in some form; of these, some 336,000 tweets were closely engaged with the question of Islam in Europe. The majority of these tended to defend Muslims and point out that ISIS does not represent Islam; some one quarter, on the other hand, attacked Islam and Muslims directly.

These tweets were geolocation (using GPS data as well as inferred locations); users from Muslim countries tended to be most defensive of Islam, while countries such as Israel and the Netherlands were most critical. The most retweeted anti-Islam tweet was by Donald Trump, incidentally.

Is it possible to predict the stance of individual users towards Islam, even if they have not previously discussed such topics? The project now focussed in on some 44,000 US-based Twitter accounts that were seen to be either defending or attacking Islam; it then examined the content of all of these users' previous tweets, their profile information, and their interaction network (retweets and @mentions). The majority of these accounts had never mentioned Islam on Twitter before.

The network information was most effective at predicting users' stances towards Islam: users with positive views followed the liberal media and Democratic leaders, discussed political issues, and were interested in music; those with negative views followed conservative media, Republican politicians, and sports such as NHL, NFL, and NASCAR. This shows the influence of network homophily and social influence.