Can Twitter reveal our mood?

Can Twitter be used to answer scientific questions about our collective mood?
26 June 2018

Interview with 

Professor Nello Cristianini - University of Bristol


Social media app icons on a smartphone


Twitter might reflect the mood of the nation, as we collectively share our thoughts with the masses, but can we use this data to answer serious scientific questions? That’s what a team at the University of Bristol wanted to find out. So, for over 4 years, they collected tweets, every hour, from the 54 largest cities in the UK. They then worked out what categories of words were being at any moment, and provided an indication into people’s moods at the time they were tweeting. Georgia Mills got the #story from study’s author Nello Cristianini...

Nello - The big surprise for us was there are indeed two different thinking modes. One morning orientated, one more night oriented. A morning mindset focussed on power, drive, achievement and that is a very strong signal. It peaks at 7am, 8am, 9am in the morning, and as the day progresses we see a change in this. And by the time you reach the late evening you start seeing a lot of he and she pronouns, male and female references. You start hearing negative emotions, swearing, and as you go on into the middle of the night, we have the moment when people talk about death. And just before the sunrise this is the moment when the religious topics peak. Then the sunrise starts and the cycle starts again.

Georgia - I think we can all relate to the existential dread in the middle of the night. But how did you pick apart the fact that these are trends happening to the same people because could it be that the people who are awake and tweeting at 6am are just more driven, and the people who are awake and tweeting at 3am are just more concerned about death and things like that? Could it be this is when different people are choosing to tweet?

Nello -  Absolutely right. Obviously this is not the same one person tweeting. There are different people tweeting and we are sampling them. One could say there are different types of people and some of them are active at night and that’s entirely possible. And it’s hard for us to correct because we anonymise. However, I would argue that isn’t this the same point. Then you can ask right away why are those people who are interested in darker topics active at night? And why are those people interested in social concerns active in the evening after dinner? In a way, we are coming back to the same question again: why would they be active at different times?

Georgia - Have you got any ideas?

Nello - Well, we have conjectures but we don’t have anything. So this study really is carefully pointing out that this happens. As for causation, that's difficult. One thing we did we tried to ask a follow up question.  Is it possible to summarise all this variation and explain it with a few factors. And we found that just postulating two hidden factors accounts for nearly 80 percent of all the variation, and they are cyclic, of course. One peaks at 6/7am the other one peaks at 3 am. They actually behave like some of the hormones we have. So although we cannot prove anything we do observe that this variation does correlate to some of our other hormones as well.

Georgia - Right. So this could be a flag that maybe it’s our hormones driving this change, but further work needed?

Nello - Further work is needed but we do have this suspicion. But for this paper, what we can report with confidence is the empirical observation that these things are cyclic. They have a 24 hour cycle and this is statistically significant.

Georgia - How many tweets would you say in total you have analysed?

Nello - 800 million tweets.

Georgia - Quite a big number. Is this going to change how we do research what you’ve done here? Is this just the tip of the iceberg?

Nello - Yeah. My hope this this; that we can start using these things in a good way. You could repeat the same study by looking at different types of data. For example, in the past, we’ve been looking at the log of the queries in Wikipedia. What are people searching for in different seasons? And we find various seasonal patterns in their queries. So data is becoming available. It’s open and if you have the right scientific question you can design a study and get an interesting answer.

Georgia - What kind of questions would you like to ask?

Nello - At this moment I’m still looking at seasonality and daily cycles of emotions and thinking patterns. Previous studies we have done show that there is an increase of negative emotion and sadness in the winter and we try to look at mental health. Many years ago we did demonstrate that you can use this to detect flu epidemics, so we are just exploring what is possible.


Add a comment