Tracking The 'Truthiness' Of Tweets
IRA FLATOW, host:
You're listening to SCIENCE FRIDAY. I'm Ira Flatow.
Let's say you're running for office and you want to smear your opponent. Where can you or one of your supporters go to spread a nasty rumor? Well, it could be the radio, could be TV, but what about the new social communities, like Twitter?
Twitter seems to be the new frontier in political campaigning and the perfect place to launch a campaign of misinformation that can move through cyberspace and into the minds of voters at warp speed, before they even know what the other side might be saying.
So as a voter, how can you tell if the information you're getting from Twitter is legit or is something that may have been posted to deliberately mislead you? You need a truthy tracker.
Researchers at Indiana University in Bloomington have come up with a way to track Twitter feeds and sniff out those tweets that might be lies. And to demonstrate how this works, they've agreed to track our tweets on SCIENCE FRIDAY.
If you follow us on Twitter, @scifri, @-S-C-I-F-R-I, re-tweet the post -truthy@indiana. Re-tweet that and take part in our experiment to see how far and wide a tweet can spread.
You can go to our website at sciencefriday.com and also do it that way, and you can see the results of the map that's being created, tracing all your tweets.
And we'll talk about that now with Johan Bollen. He is associate professor in the School of Informatics and Computing at the Center for Complex Networks and Systems Research at Indiana University in Bloomington. Welcome to SCIENCE FRIDAY.
Mr. JOHAN BOLLEN (Indiana University): Oh, hi, Ira. I'm a big fan of the show.
FLATOW: Thank you. How did you what was the genesis of this idea?
Mr. BOLLEN: Well, I think all of us have been noticing that the political debate and the political discourse in the United States over the past, well, say, five to six years has become very polarized, very partisan.
And so one of my colleagues, Filipo Menzer(ph), actually, came up with the idea for this website, read a newspaper article about how a group of people in Massachusetts tried to actually change the outcome of the election by creating a so-called Twitter bomb, where they set up a bunch of accounts and sent out thousands of tweets on the morning of the election, right, hoping that it would be re-tweeted and generate enough travel, traffic so it would be picked up on Google.
And the tweet in question actually contained a URL that pointed to a Web page, as I understand it, that looked like a real news article but actually contained a false news story.
And you know, of course we don't know whether that affected the outcome of the election, but it could very well have done that or at least have been a factor.
So when we heard about that, and we put that together with sort of our discomfort about some of the material that we've seen distributed online, like most other people recently, we figured to provide some kind of an antidote to that development.
FLATOW: Well, how can you tell truthiness - our hats off to Stephen Colbert for that word, how can you tell the truthiness of a tweet? Now, we have a tweet going on our website. It looks like, to me, I'll describe it. It looks like the head of a dandelion seed, you know, with all the lollipops coming out of it, and with a center in it, that we started out as the tweet. We want people to re-tweet. How do you know if that's a real tweet or if it's some campaign smear?
Mr. BOLLEN: Well, you don't actually know because it's really difficult to make an objective decision on whether something is actually true or false. But you can definitely look at the pattern, you know, of how this particular meme spread through the Twitter network.
I mean, you see one big central node right in the middle of that line that you mentioned and then a whole bunch of other nodes that have re-tweeted that same tweet.
So these are users that are essentially following marching orders, and the particular structure of the graph and its evolution over time is actually an indication that something may be going on there that is not entirely natural. And in this case, of course, we know there isn't because we've actually started this.
FLATOW: Because it has a central spot and everything's branching out from the middle, that's an unnatural look.
Mr. BOLLEN: Well, I wouldn't say it's an unnatural look, but it's one of the many factors that you could look at in terms of how a particular tweet gets started and how it is distributed.
I mean, there's a number of factors that we're trying to look at in this architecture, you know, among other things the number of tweets per user - you know, the sentiment that - the content of that tweet contains.
So we're looking at a whole bunch of different factors, and in the future we're hoping that we can actually automate, you know, set up automated methods that can look at all of these different features and then provide some kind of an assessment of how truthy that particularly meme is.
FLATOW: And are you thinking of setting up a site that will evaluate some of these tweets and say be aware, potential...
Mr. BOLLEN: Yeah. I think that's more or less the intention of this project. At this point we've only just started to collect the data and to design our methods. But we're hoping that in the future this could lead to sort of a generalizable method, not just for, you know, the upcoming elections but for a number of other applications where we would be able to track these memes, look at their features, look at their, you know, how they're being communicated online, you know, who actually started it - you know, where it came from, look at synchronization, apply epidemiological models to how these tweets spread and try to sort of tweak a signal from all of these different features and try to determine whether, you know, something really suspicious is going on.
FLATOW: Right. Give me an example of how, if I were to start one, what I would do, a mechanism in my objectivity here?
Mr. BOLLEN: Oh, if you wanted to start a truthy meter?
FLATOW: Yeah, if I want to deliberately spread misinformation, what would I be doing?
Mr. BOLLEN: Well...
(Soundbite of laughter)
Mr. BOLLEN: Perhaps I shouldn't be telling you. But I would say that you would probably have a group of people. You would agree to set up a Twitter account under some false pretense. You would design you would craft a particular tweet, a particular salient, emotional content, something that people would like to pass along, that looks kind of true but perhaps isn't.
And then you could start to coordinate spreading that meme and sending it to your followers and so on. You know, it's really as simple as that. Since we don't have a very precise machine definition of what it means to be - for a meme to be truthy, it's really difficult to tell how to best do that.
But the fact is that people are doing this. There's no doubt that both political parties or any political groups are actually using Twitter deliberately to spread misleading and deceiving information.
FLATOW: And one other interesting part of this is that you mentioned at the beginning that once this starts to spread, it makes its way into the Google search engine, and that adds another layer to it.
Mr. BOLLEN: That adds a really interesting feedback loop because once it generates enough activity, and it's actually listed at the top of those search results, then people may actually click on that tweet and may start to re-tweet it themselves, and that generates more traffic, and it increases the ranking of that tweet on Google, and so you get a really nice positive feedback loop where from very humble beginnings, so to speak, you can launch a meme into the public consciousness and have it propagate.
So it's really a little bit like studying flu epidemics, and one member of our team is one of the world's leading experts in how diseases actually spread from person to person, and the process in which these tweets spread from person to person may be modeled along the same lines.
FLATOW: Is there anything illegal about doing that?
Mr. BOLLEN: No, I don't think so. I mean, it's really difficult to well, let me put it this way. I think there's laws on slander and so on that would stop you from posting things that are, you know, illegal or hate speech. But I think it's perfectly viable to come up with something that is truthy, you know, just truthy enough then to not be illegal and to make it popular enough to affect political outcomes.
FLATOW: And so are you going to be watching for this, this election cycle?
Mr. BOLLEN: Yes, we're very much hoping to do that, and hopefully we can continue this effort past the elections because there will be more elections, and there will be more events where it will be necessary to monitor the truthiness of the memes that are circulating on Twitter.
In fact, we're fully anticipating releasing our data and our software to the public so other people can actually build on top of what we've done.
FLATOW: Well, we have a video up on our website, at sciencefriday.com, of this re-tweet that we started. And if you want to go look at it, you can go to, as I say, go to sciencefriday.com and watch it spread and watch the re-tweets happen and get sort of a demonstration of what you're talking about.
And I understand you're going to be updating it for us as we go along.
Mr. BOLLEN: Yes, yes, because apparently that network is still evolving, and, you know, on our website we actually showed the amount of activity over time that a particular meme receives, and you can actually really see that there's a first wave where that tweet was actually launched, and we can see sort of the spike in activity with regards to that meme and then a second spike of activity that seems to correspond to the network developing additional branches, and so on. It's a really nice, it's a really nice animation.
FLATOW: And how soon will you have other, other of these up there? If you find something suspicious, will you be posting things yourself as the weeks go on?
Mr. BOLLEN: Well, the website automatically updates. So whenever a meme achieves a certain level of activity, we catch that, and we update the website to include that meme. So it's constantly refreshing in real time, and you know, if anyone launches a truthy meme that gets enough traction, we'll be able to find it in our data.
FLATOW: You know, this is always sort of a spy-versus-spy world. Do you think people are now looking already at your truthy ideas and saying, gee, how do we get around this thing?
Mr. BOLLEN: I bet they are. You know, that wouldn't surprise me. But at the same time, hopefully we've increased the cost of doing these things, just enough to make it a little less likely.
FLATOW: You mean the real the physical or mental, or is it actually monetary cost?
Mr. BOLLEN: Well, I wouldn't say monetary cost, but you know, if right now it's really easy to set up, you know, 10 Twitter accounts and start spreading these memes and hoping they go viral, and they get a lot of traction.
But it may very well be that in the future, when people actually catch those kind of patterns, like the dandelion that we're seeing with the meme that we created and that we launched deliberately, you know, once people notice that that's one of the features we're looking at, they may have to involve a lot more people in spreading these kind of memes.
FLATOW: Is there any second generation or third generation to this? Are you thinking ahead or just trying to get this, this stage settled?
Mr. BOLLEN: Yes, well, we're first trying to get this stage settled. So we're definitely looking at all of the different features. Actually, users can go to the website right now and, for example, report whether they think a meme is particularly truthy.
There's a whole bunch of other features, like the number of users per tweet, the growth of the meme and so on. And we're definitely looking to generate automated methods that would allow us to look at all of these different features and then provide some kind of an assessment to the truthiness of the meme.
FLATOW: Well, if you'd like to participate, we'll be going all day, and for the next few days you can go to our website at sciencefriday.com, and you can tweet us @scifri, join our Twitter group and then re-tweet truthy@indiana and see how well that works. And it's all there on our page at sciencefriday.com.
I want to thank you for taking time and for helping us out with this crowd-sourcing today.
Mr. BOLLEN: No, our pleasure entirely.
FLATOW: So it will be interesting to watch for the weeks and watch the weeks ahead of how it all works out. Thanks again, Johan.
Mr. BOLLEN: No problem. Thanks a lot.
FLATOW: You're welcome. Johan Bollen is associate professor in the School of Informatics and Computing at the Center for Complex Networks and System Research at Indiana University in Bloomington.