Using our social media analysis tools, we attempt to find out whether David Cameron, the Prime Minister, and Theresa May, Home Secretary, were correct when they claimed that social media, primarily Twitter, Facebook, and the Blackberry Messenger service, were used to organise and incite the riots.
In order to test this theory, we used our social media listening tool to collect microblogs (Twitter and Friendfeed) containing the words ukriots, londonriots, looting, or rioting. A set of four hundred posts was then classified manually, in six categories – 3rd person report, 1st person report, rumour, incitement, comment, and not applicable. This was used to train our classifier, and the remainder of the data was subsequently fed through.
The output data was separated according to classification; the category with highest probability was assigned. Note that, as the language used in the categories was fairly similar, the “winning” probabilities were actually quite low in some cases. The following graph, of hourly counts for each category, was then produced.
This shows that, from the start of the monitoring period, there were more third person reports and rumours than there were posts classified as “incitement”. The peak for “incitement” around 9:00am on the 8th is for the following tweet:
Middle class looters in Windsor steal cream, strawberries & meringues from Waitrose and cause a right Eton mess. #londonriots
This was retweeted over a hundred times, and appears to have been wrongly classified in the training set. The next major peak for this category, at 97 hours, was for the tweet
#chavbands JLS. ( just looting sainsburys )
Which was also retweeted more than a hundred times, and also appears to have been wrongly classified in the training set.
Tweets classified as rumour are, in some instances, very similar to those classified as third person reports; the criteria in the training classification was whether they included words like “just heard”, or asked whether it was true that a particular store or area was hit by the riots.
This quick and simple classification of tweets collected demonstrates a number of things:
- Care is needed when using social media to find the instigators of criminal activity. Law enforcement agencies need to determine whether a post is humorous or serious in intent.
- If twitter was being used to instigate or organise the riots, the people posting were not using any of the keywords used in this search. A wider search would be needed to confirm or refute the government’s claim.
- Sufficient categories are required to take into account all the variability; in this case, a category of “Humour” should have been included.
- A larger training set is needed, in order to properly separate out the categories – there appears to be an overlap between third person reports and rumours, and the probabilities used to assign posts to classes were in some instances very low.