Scientists say Twitter data unreliable to understand popular emotions

Twitter is an unreliable witness to the world's emotions, and assuming that the micro-blogging site accurately reflects real life may be risky, scientists say.

With over 300 million monthly active users around the globe sharing their thoughts in 140 characters or less, studies based on Twitter data are "particularly alluring" to researchers and the media, said sociology expert Eric Jensen, from the University of Warwick in the UK.

However, he cautions against this "big data gold rush," pointing out that there is no evidence that social media content shared on Twitter is a truthful reflection of how its users feel.

Twitter users have developed their own unique cultural behaviour, conversations and identities, which shape the ways in which they present their views online.

Social convention, power relationships and identity influence online conversation just as much as off-line interactions, but in ways that are not yet fully understood.

In a study published in the journal PLoS One Jensen also highlights the problems of drawing broader conclusions from a sample of Twitter users.

It has been proven in several studies that Twitter users are not representative of the general population. In just one example, men are much more likely to use Twitter than women.

Prolific users who tweet many times a day may be over- represented in any sample dataset.

"Twitter users present only one side of themselves on social media, shielding their true feelings for good reasons, such as professional reputation," said Jensen.

"There is clearly a large gap between what people post on social media and how they really feel, but how exactly people manage the relationship between their offline and social media identities is still being uncovered," he said.

"When researchers find themselves with easily accessible data, there is a temptation to apply those data to interesting research questions and populations - even when there are limitations in the representativeness of the sample, he said.