We're hopeful that our instrument will be useful for governments and policy-makers, for analysts of all kinds, and for individuals wanting to know how their society is faring. We hope that that the hedonometer and its future iterations will become an essential dial on the dashboard of society.
Professor Peter Dodds
Mathematicians from the University of Vermont (UVM) have created an instrument to measure the happiness of large populations in real time. With help from The MITRE Corporation, the UVM scientists have developed the hedonometer, which shows that the saddest day during the last five years was Monday 15 April 2013 – the day of the Boston Marathon bomb attacks.
As of yesterday, the findings of this ‘happiness sensor’ are freely available on the web. At the moment, updates are on a daily basis, but the rate is soon to be upgraded to every hour and, eventually, every minute. The measurements are currently taken on the basis of Twitter posts in English, but there are plans for expansion there, too.
The hedonometer is the brainchild of UVM’s Professor Peter Dodds, Director of the Vermont Complex Systems Center, and Chris Danforth, Associate Professor in the College of Engineering and Mathematics’s Department of Mathematics and Statistics. Together, they run the Computational Story Lab.
"We started to think about measuring the emotional states of large populations at the end of 2006," Professor Dodds told ScienceOmega.com. "We were initially inspired by the site wefeelfine.org, and it took several years for us to develop and then refine an instrument."
The resulting instrument uses technology developed by Brian Tivnan, Matt McMahon and their colleagues at MITRE to effectively compile and average the emotional state of large populations through the words contained in a random sample of tweets. To do so it draws on a database of around 10,000 words which have been rated by the paid volunteers of Amazon’s Mechanical Turk service for their relative happiness on a scale of 1 to 9 (the happiest). While 'happy' receives an 8.30, 'hahaha' comes in at 7.94. Neutral words such as ‘and’ and ‘the’ score around the halfway mark at 5.22 and 4.98, with the emoticon ':-(' and 'war' near the bottom of the scale, rating 2.36 and 1.80 respectively.
"We're measuring how people present themselves by their writings in aggregate, so we're not looking inside anyone's head," explained Professor Dodds. "Happiness/sadness is arguably the main emotional dimension for people so we're starting there, but we will be adding other key emotions as time goes on, such as anger and fear."
The team explain that, while they are not attempting to define happiness, measuring the independently-reported positivity or negativity of the words people use in their tweets can operate as a finger on the pulse of society’s emotional state in the same way that Gross Domestic Product (GDP) acts as a remote measure of the state of an economy.
"While we're clearly focusing on in-the-moment happiness as expressed by an online population, we've shown the signal coming from our instrument to correlate well with traditional survey measures of a variety of indices of well-being," added Professor Danforth.
In February, for example, the researchers were in the news with a study of geo-tagged tweets from i-Phones in which they mapped the happiest cities in the United States. Further investigation highlighted the socioeconomic factors that contributed to their results as well as the correlation between food-related word usage in tweets and obesity.
The initial focus of the project was weblogs. In a study published in 2009, the pair showed that happiness as expressed in song lyrics had decreased over 50 years, and that happiness in blogs had grown over a four year period.
Sources beyond social media
"Presently, it relies upon a random sample of ten per cent of all public tweets – a collection of roughly 50 million messages each day," Professor Danforth related. "Future iterations of the site will also measure sources beyond social media, including Google search terms and traditional news media."
Although there is too much variation in the way that an individual uses a particular word – take ‘sick’, for example, which can mean ‘unwell’, ‘vomit’ or ‘absolutely amazing’ – to gain understanding of a small group or sample. However, with a sample of 50 million messages per day, it’s the overall effect that paints a picture. Twitter may not provide a representative sample of the population, but it has become a potent collective voice.
"We're showcasing Twitter as an example for our working prototype, but we can use our instrument for any large enough text source," Professor Dodds went on. "For example, we have 20 years of the New York Times, data from the Google Books project, and music lyrics. We'll be adding these and other feeds along with 12 languages besides English in the next few major updates to the site."
MITRE has been developing the infrastructure required to make daily updates to the instrument measurements, and their code is capable of producing updates every hour, and even every minute.
'An essential dial on the dishboard of society'
"As an analogy, consider measuring the temperature of a gas," suggested Professor Danforth. "Temperature is not a meaningful measure if only a few hundred molecules are considered, but when billions of molecules interact, temperature tells you something about the state of the system. Since our instrument is based on text, we're in the process of experimenting with the size of the text sample required for us to be comfortable with the output."
"As we all know, social media has taken off enormously in the last few years; a tremendous amount of data is being delivered in real-time," said Professor Dodds. "This certainly couldn't have been done five years ago. We're also using Amazon's web services, which have been harnessed by our collaborators at The MITRE Corporation."
While work is underway to explore the potential of monitoring two-word expressions and of using brain imaging techniques to back up the remote sensing approach, the team are keen to share their findings so far with the wider world.
"We hope that the general public will enjoy seeing the emotional state of the internet, and that journalists will look to the daily word-shift as a scientifically defensible representation of public sentiment in response to global events," Professor Danforth responded.
"We're hopeful that our instrument will be useful for governments and policy-makers, for analysts of all kinds, and for individuals wanting to know how their society is faring," added Professor Dodds. "We hope that that the hedonometer and its future iterations will become an essential dial on the dashboard of society."
"Ultimately we would like for policy makers to think of the hedonometer as an indicator of societal health and collective performance to complement more traditional indices like GDP or Consumer Confidence," concluded Professor Danforth.