Information Cascades in Twitter
Twitter is a great tool for examining how information travels through a network and can be particularly effective for spreading information in a crisis. For instance it has been shown that Twitter can give information about epidemics and disease dynamics well before official health reports can do the same (http://www.ajtmh.org/content/86/1/39.abstract). Researchers from Rochester Polytechnic Institute and Rutgers University have examined how information about a crisis is spread through Twitter as an information cascade in a recent article titled Information Cascade in Social Media in Response to a Crisis: a Preliminary Model and a Case Study.
In this article the authors examine a specific instance in which a robber was spotted on campus and three warnings were sent out and finally an all clear message. In this case the set of Twitter users tweeting, retweeting and replying before the event was small, but following the first warning all three categories skyrocketed and grew larger following the second warning. Tweets and Retweets began to diminish when reaching the third warning and then after all clear, though both were well above the pre-warning values. The authors focused upon retweets, and found that information began to spread (the graph of retweets and users becomes more connected) after the first warning and continued to until after the third warning, at which it began to decline.
This article is particularly relevant to the class because it is a case study for how information cascade actually occurs in a modern social network during a real crisis. In this case it shows that information cascade isn’t just some theoretical topic in Network and Information theory, but rather something real, observable and important in the real world. Information Cascading is important to spreading news, particularly in the information age and is especially useful in crises.
In general a Twitter user is in one of a few states, according to the article–has not received any message, has received but does not believe, has received and is uncertain and finally has received and does believe. The action of each user is simple from this model and follows principles of information cascade. Those that have received the message and are uncertain of its value look at neighbors to see what seems to be true–the situation of someone who has drawn a marble and everyone else has said the urn is majority blue will say the urn is majority blue regardless of their information as their neighbors have a belief in the message. If looking at neighbors confirms the users belief the message is spread. In the case of a crisis, as authorities began to tweet about the crisis, general users see it and began to tweet and the message spreads–users who receive the tweet from other general users may search out neighbors or authorities for “confirmation” and then pass the message along–thus cascading the information through the network. Of course, cascading effects in a crisis may not always be accurate, for instance during Hurricane Sandy information went viral (spread through the Twitter network) that the stock exchange was flooded and NYC was turning off all power, but was in fact false and was tweeted by someone masquerading as an authority (http://www.latimes.com/news/nation/nationnow/la-na-nn-twitter-social-media-sandy-20121105,0,1510592.story).
Source article: http://www.cs.rpi.edu/~magdon/ps/conference/InfoCascadesSWDMwww2012.pdf
From the diaries of Potatis