Identifying Events from Twitter Bursts

Reykjavík.
The next speaker at ECPR 2011 is Andreas Jungherr, whose interest is in using Twitter data to detect events by identifying sudden bursts of activity in the continuing stream of updates. Such research is especially straightforward on Twitter, due to its convenient API access formats; additionally, the short format of Twitter messages means that key themes in messages can be more easily identified.

Twitter itself does some of this, of course, with its ‘trending topics’ (also broken down for specific geographical regions); further, it is possible to identify the links which are shared as part of tweets, of course, as well as identifying hashtags, @replies, and retweets. And tweets are exactly timestamped, allowing for close analysis of temppral developments.

Andreas notes the ‘Stuttgart 21’ protests (against the rebuilding of the central train station) as one example of an event to analyse: he tracked the messages of some 80,000 German Twitter users, and examined the relative peaks and troughs in the presence of the #s21 hashtag. Some 12,000 users generated over 140,000 messages between 1 June and 19 October 2010; of these, 76,000 were retweets, and 17,000 @replies.

Within these tweets, specific key terms tend to show some very significant peakiness: they are unique to a very limited timespan, occurring mainly within a few hours. Twitter events (at least in this case) mapped to offline events, but the opposite is not true.