Twitter had more than one hundred million active accounts in October of this year. The network therefore comprises an increasingly large data warehouse and offers academic researchers a huge field for behavioural studies.

A group of mathematicians and statisticians from the University of Vermont* (Burlington, USA) has drawn up a picture of the topology of social networks by studying the way in which users interact and observing how information is communicated via this online platform. The work poses both theoretical and computational challenges. Several recent studies have examined social networks starting from key words used on Twitter and from the behaviour of Twitter “followers”. The team of researchers at Vermont criticises this approach, which is based on “follower” behaviour only. The team proposes a different model, based on a sample of reciprocal replies. The Vermont study thus offers a more dynamic view of the network, detecting reciprocal changes at the level of days, weeks and months, going all the way through to the phenomenon of “unfriending”.

Happiness breeds happiness…?

The Vermont model enables the relationship between happiness expressed and Internet user activity, in this case via the “tweet” maximum of 140 characters, to be tested. Users’ average happiness scores are positively correlated with nearest neighbours’ average happiness scores one, two and three links away. These correlations decrease with increasing path length between correspondents. The results provide evidence of a social sub-network structure within Twitter. The analysis of 40 million short message pairs – from a database of 100 million tweets posted between September 2008 and February 2009 – reveals apparent mood “contagion” up to three levels.

Other similar studies

Several other studies have already used Twitter as a means to detect Internet users’ changes of mood over a day or a year, by watching out for certain key words that help to reveal happiness, depression, or even intoxication. Scott Golder and Michael Macy of Cornell University have analysed the tweets posted in 84 countries over two years. Results show that tweeters are generally in a good frame of mind when they wake in the morning, but that their mood sinks as the day wears on. Exchanges are also more positive during the weekend than in the week “probably due to distinct sleep and biorhythmic patterns,” thinks Scott Golder (Science magazine, Tweets Reveal the Role of Circadian Rhythms on Mood). In fact, “the most popular users - as measured by ‘follower’ count -  may not be the most influential in terms of information cascade, as indicated by the number of messages that are re-tweeted,” the Vermont statisticians’ report points out.

* Catherine A. Bliss, Isabel M. Kloumann, Kameron Decker Harris, Christopher M. Danforth, Peter Sheridan Dodds,