Monitoring Moods with Twitter

An analysis of over half a billion tweets worldwide has confirmed that we’re all in a better mood in the morning...
04 October 2011

Interview with 

Michael Macy, Cornell University


Twitter bird logo


An analysis of over half a billion tweets worldwide has confirmed that regardless of country or culture, we’re all in a better mood in the morning.  Speaking with Chris Smith, Michael Macy, at Cornell University, has analysed the messages posted on Twitter by 2.4 million people from 84 different countries to find out more...

Michael -  We were looking to see how people's moods change over the course of the day, over the days of the week, and over the seasons of the year using Twitter which allows us to monitor people's expressions in real time at a global scale across diverse cultures, which is something that scientists have not been able to do in the past.

Chris -  Talk us through the method.  You presumably didn't read half a billion tweets individually?

Michael -  That's right.  It's all a process using computers.  In fact, you wouldn't want to try this on your desktop computer - we used Cornell's Centre for Advanced Computing, a large cluster of powerful supercomputers, and what we did is basically for each user on a given hour in a given day, like let's say, Tuesdays at 9am to 10am, we would take all the messages that they wrote during that time period and throw them together kind of as a 'bag of words'. Then the computer sorts through that bag and identifies all the words that are positive words like 'awesome', 'fantastic', 'incredible', 'hilarious', and also finds all the negative words like 'unhappy', 'depressed', 'embarrassed', 'anxious', 'afraid', and so on.

Chris -  How does it handle sarcasm?  So if I said, "Brilliant!  Just got fired."  How would it cope with that?

Michael -  It would get it wrong, and one of the advantages of having such a large number of messages is that the errors will tend to cancel out at very large numbers.  So, you get a little noise with people using sarcasm, or for example saying, "Good morning" in the morning just as a ritual but not as an expression that they're actually feeling good.  But when you have this large a set of observations, in this case over a half billion messages, then the errors tend to cancel out. The reason that we know we're seeing the errors cancel out is that we get a very robust pattern -  it holds up.  We see the same thing each day of the week, and we see the same thing across very diverse cultures. 

Chris -  And since you say you saw these patterns, what patterns did you see?  What did the analysis reveal?

Michael -  Well basically, we found that people are happiest in the mornings and then it's sort of all downhill from there!  But then, in the evening, there's a rebound in positive feeling, right on up to bedtime with a second peak in the late evening.  And we see this pattern every day of the week so it's not just work that's doing it, but it certainly fits with what we think about work -something that makes us tired, it gets us stressed out, the frustration of commuting, and so on.  We actually see the same pattern on the weekend, so these two peaks with a trough in the late afternoon seems to be something that's kind of built-in to human body rhythms and it's independent of the day of the week or whether you're at work or not.  On the other hand, people are also happier on the weekend which does suggests that there's something going on with work, but the peak on the weekend is a little bit later.  It's about one and a half to 2 hours later than it is on a weekday which suggests that people are sleeping in, whereas on the workday, they're getting woken up by an alarm clock.  On the weekend, they're perhaps sleeping until their body is ready to wake up, and so it could be that this weekend effect, the elevated mood on the weekend, is actually a response to having sleep that is not interrupted artificially by an alarm clock.

Chris -  It also tells me that the people you analysed in your study obviously didn't have kids because they wouldn't have been afforded the luxury of a 2-hour lie-in! What about one other quite important thing which I guess you may have been able to test this with this sort of data set - seasonal affective disorder.  The whole idea that as we shift through the seasons, people at very high latitudes are being exposed, not just to an absolute reduction in day length but to a changing day length.  Do you see this mirrored in the moods indicated by your analysis?

Michael -  Yes, we do.  But it's actually not the absolute amount of daylight that you're getting or how long the day is in absolute terms.  It is the change from what you're used to.  So, what's happening is that as the days are getting longer, moods are improving and as the days are getting shorter, then moods are not so good, and it's the change in the daylight, not the absolute length of the day that seems to be associated with changes in mood.

Chris -  So what's the bottom line with this study?  Why is this important?  What have you flushed out from doing this that we didn't know previously from other sociological experiments on small groups?

Michael -  Because we had such a large set of observations from all over the world, we're able to have more confidence that we're nailing down when in fact it is that people are in the best mood.  We also had some surprises in the results - finding the same pattern on the weekend and finding the same pattern across very diverse cultures was something of a surprise. You would think a country that's collectivist for example, or has a different religion from the US, UK, or Australia, you would think that perhaps they would have different patterns of when people are in a good or bad mood.  But in fact, we found very similar patterns all across the globe - whether it's India, or Africa, or the UK or the US, Australia - very similar patterns.

Chris -  What are the implications of that? 

Michael -  Well one of the things we'll want to do is just to break it down a little bit into more detail of the different groups within this large population in different ways.  So for example, one of the things that we looked at that we might want to investigate further are different chronotypes.  A chronotype is a person who's active at a particular time of day such as a night owl or a morning person.  Another one of those surprises was that we found out that night owls are different from everybody else in a sort of peculiar way.  Everybody else has their second peak when their moods rebound in the evening, and you would think that certainly night owls will too, but in fact we found the opposite.  The night owls just have that peak in the morning like everybody else, but they don't have that rebound in the evening.  They're basically the group that's the least happy in the evening and yet they're the group that's the most active.  So that was a bit of a surprise and we'll pursue in more detail those kinds of breakdowns of who in the population is reacting in what way. 


Add a comment