Advertisement
Advertisement
Filippo Menczer: Since 2009, we've been working on studying the diffusion of memes and information on social media, especially on Twitter because we were fortunate to have access to a lot of data from Twitter. We have several grants supporting this research, and some of it is very much towards building systems. For example, to visualize or analyze how information is spread from one person to another, studying what makes memes go viral, and the competition among different memes for the finite attention of users. Is it the structure of the network or is it another factor that make these memes go viral? These are general things we wanted to address with Truthy about how social media affects the spread of information, and also the spread of misinformation.What else have you worked on since starting the project?
We also built models where we tried to replicate very simplified scenarios in which people use social media. We look at the predictions of these models, like which memes go viral, or how two different opinions compete with each other, and then we see if we can reproduce some empirical patterns from the data. We've looked at if people can distinguish real humans from bots on Twitter, and we've had some success with that.More recently, we've been studying whether you can detect whether a meme or a hashtag is being promoted by advertisement, versus information that is being spread organically.
Advertisement
That's a very interesting question. A lot of people have looked at this issue and found that different factors may affect the virality of a meme, such as time of day, color, and pictures included. But we were curious to know if that was, in fact, a causal relationship.We tried to explore this question in a theoretical way by building some models that simulate people using Twitter or Facebook, where something pops up on a user's screen and you can retweet it, skip it, or create new content that would go to your friends or followers.What did you find from the experimentation?
What we found in this model was that something almost always went viral when there was an underlying structure of a social network where the users had a finite amount of attention. Yet, in this model, the things that went viral were no different than the things that did not go viral.So sometimes it is inevitable that things will go viral due to the structure of the social network and because of our finite attention. To say, "This is the thing that makes a meme go viral," you have to prove that it's not just a coincidence. The things that go viral may have some things in common, but that doesn't mean they're the reason why they went viral.One feature that we did find correlated with virality is whether a meme, in its early stages, spreads through multiple communities.Meaning multiple platforms, like from Twitter to Facebook?
Not exactly. For example, say you have high school friends, fellow journalists, and people who like tennis who follow you on Twitter, and these are your communities, or commonalities. If a meme, very early on, spreads through many of these communities, it's much more likely to go viral.
Advertisement
Yes. If [a meme] spreads to 10,000 people very early on who are all in one community, then it's less likely to go viral. But if it goes through 100 people who are all from different communities, and they interact with it, it's much more likely to go on and reach a million people.
In 2012, we worked on a research paper where we looked at the structure of the networks formed by people tweeting and retweeting about words associated with the 2010 midterm congressional elections. This is a slightly dated analysis, but I can summarize something interesting that we found. I assume that it still holds true.We looked at diffusion networks through Twitter, where a "node" is a user and a "connection" means that a particular meme has been transferred from one user to another person, either through mention or retweet. We did this for all hashtags that were used to talk about politics in 2010.What'd you find?
We found that there was huge polarization and segregation between the two big communities. The liberals retweeted the liberals and the conservatives retweeted the conservatives, but there were very few liberals who retweeted conservatives, and vice-versa. We use this term "echo chambers" to describe how people are exposed to information that is aligned with their preexisting beliefs. We're very interested in the fact that these echo chambers still exist. For example, if you have a friend on Facebook who has a different opinion, you can just unfollow them, right? So that makes it very easy for those echo chambers to form, and these may very strongly effect the conversation about politics on social media.
Advertisement
There are some colleagues at Indiana University who found that you could use Twitter as a relatively accurate poll. There was a clear correlation between the number of tweets about the politician and the number of votes that they received. So that would suggest that yes, if there are memes about candidates that are more popular online, then it's fair to assume that candidate may be more likely to win.For more on the Truthy Project visit their website.Follow Gabriella on Twitter.This interview has been edited for length and clarity.