杏吧原创

America, are you happy? The emotional words contained in hundreds of millions of messages posted to the Twitter website may hold the answer.

Computer scientist at Northeastern University in Boston and colleagues have found that these 鈥渢weets鈥� suggest that the west coast is happier than the east coast, and across the country happiness peaks each Sunday morning, with a trough on Thursday evenings. The team calls their work the 鈥減ulse of the nation鈥�.

To glean mood from the 140-character-long messages, the researchers analysed all public tweets posted between September 2006 and August 2009. They filtered them to find tweets that contain words included in a psychological word-rating system called 鈥� a low-scoring word on ANEW is considered negative, a high-scoring one positive. They also filtered out tweets from users outside the US, and also from those in the US who did not include their exact location 鈥� for example, their city 鈥� in their Twitter profile.

That left 300 million tweets, each of which was awarded a mood score based on the number of positive or negative words it contained. For example, 鈥渄iamond鈥�, 鈥渓ove鈥� and 鈥減aradise鈥� indicate happiness, whereas 鈥渇uneral鈥�, 鈥渞ape鈥� and 鈥渟uicide鈥� are negative. 鈥淒entist鈥� is fairly neutral.

Finally, the researchers calculated the average mood score of all the users living in a state hour by hour and so created a timed series of mood maps. They morphed the maps so that the size of each county reflected the number of Twitter users living there (see maps, right).

Simple or sophisticated?

鈥淭he visualisations are amazing and I think it is absolutely fascinating to see the nation鈥檚 mood vary in near-real time,鈥� says of Indiana University in Bloomington, who was not involved in the work but who is one of several other researchers using Twitter as a tool to track the public mood.

He thinks Twitter and similar sites will spawn 鈥渟ophisticated systems鈥� for mood tracking 鈥� although paradoxically, the maps produced by Mislove鈥檚 team are made by a remarkably simple method.

For example, Mislove admits that subtleties are lost because the method takes individual words out of context: if someone tweets 鈥淚 am not happy鈥�, the team鈥檚 method counts the tweet as positive because of the word 鈥渉appy鈥�. 鈥淚t鈥檚 a very naive and simple approach,鈥� he says.

Nevertheless, he thinks there is good reason to suspect the results reflect real mood swings 鈥� at least among Twitter鈥檚 users. For one thing, the national weekly and daily trends are also seen in individual regions, something you would not expect if the variations were random.

This is particularly marked when the daily mood maps for the west and east coasts are compared: the west coast mood follows the same pattern as the east, with the 3-hour time-zone delay, indicating that each coast experiences the same time-dependent swings.

Data overload

Mislove speculates that a signal shines though because the sheer abundance of data means that occasional misinterpretations are lost in the crowd. at Carnegie Mellon University in Pittsburgh, Pennsylvania, agrees. With colleagues, he recently used a to determine whether Twitter mirrors conventional opinion polls. 鈥淭he volume is massive, so the subtle stuff kind of washes out,鈥� he says.

Because Twitter data is publicly available, Routledge says mood can be sampled more quickly, simply and cheaply than using traditional polling tools 鈥� albeit more crudely.

at University College London, who also crowdsources data through Twitter, agrees. For all of the problems with decoding the data, 鈥淭witter offers researchers a unique, live data set that changes by the minute鈥�, he says.

Read more on the