Now you can read us on your iPhone and iPad! Check out the BTRtoday app.
The government is watching you.
Of course the government is watching you, who can possibly be surprised by that in this day and age? The issue is assuredly more nuanced than that, because watchfulness can take many forms and bear many implications.
A new study out of Arizona State University, commissioned by the Office of Naval Research as part of a larger project on the role of social media in crises, has found a model than can predict, with 70 percent accuracy, whether your next tweet will be in the name of protest.
On first blush, this is freaky. What does it mean that a branch of the military commissioned this project? Is the government going to be monitoring all of our tweets in the future, ready to lock us up the second we step out of line or tweet support of political protest? Suhas Ranganath, primary author of the study, doesn’t think so. At least, not using his research.
BTRtoday spoke with Ranganath about the study and the implications of the results.
BTRtoday (BTR): What did you find?
Suhas Ranganath (SR): We were focusing on people who might express protest in the future. This is scientifically or theoretically similar to people expressing interest in any topic.
We found two things. What helps to predict [protest tweets] is if the person talks to some other people who expressed interest in protest before and if that person has other interactions related to protest. We tried to model interactions and from there we tried to predict if they would protest or not.
For the data, we pulled from the protest that happened in last year’s Nigerian elections. There were protests that the election was rigged and so we pulled data from the protests there and modeled it.
The main thing we found was that for whom we interacted, we were able to predict their future interest [in protest].
BTR: Could you clarify how you applied Brownian motion theory?
SR: This is used for modeling interactions. Brownian theory is basically any fluid particles, say liquid water. Even a single water particle interacts with another particle and every interaction affects its movement. Brownian motion models how a water particle moves in a space and how collisions affect its movement.
We used the same model. One user is a particle and the space is how his interests change; if he interacts with another person (particle) his interests will change in a certain way. So we looked at how those interactions will change the user’s interests depending on how the interaction went and the type of particle he interacted with.
BTR: That was Nigeria–is this study applicable in the United States?
SR: I mean, it’s just user data. Theoretically it’s the same thing but we’re not allowed to collect U.S. people’s data. There are different guidelines, so I’m not sure. But theoretically, it’s the same model.
BTR: Not all social media posts are equal and many, especially Twitter, can have unclear intentions. How did you account for the potential ambiguity of posts with regards to protest?
SR: First we chose some keywords and hashtags relating to the protest. Then we gave it to Amazon Mechanical Turk, which is basically a code-sourcing vector, to determine if the tweet expressed protest or not. If it was ambiguous we just removed it. We only kept data clearly pertaining to the protest.
BTR: The study was commissioned by the Office of Naval Research. Has this colored people’s reactions to the study and to your findings?
SR: It’s not–they didn’t commission this exact study. It was under a broader project of studying the media during crisis. Usually research funding is in the general field, like studies in general on social media during crisis situations. So you have to realize that we researchers are given some level of freedom in what project to undertake.
It’s not exactly that the Office of Naval Research called us and told us to find people who protest; they had no specific intentions. The funding was basically for [projects on] how to analyze social media during crisis. Given my interests, I focused here.
My previous work was about how, scientifically, people interact during disasters like Hurricane Sandy. In that line [of work] I found this study very interesting so I wanted to work on it.
The military is not exactly focused on finding protesters.
BTR: They didn’t approach the broader commission with any sort of aim?
SR: No, the goal was to find ways to analyze social media during crisis. Usually research funding is awarded for a period of two or three years and with a broader goal; it’s not for a specific project.
BTR: What kinds of “real world” applications can your findings have?
SR: The model can be for anything. It can be for advertising because it’s a model of data to predict people’s actions. Say, for politics, it can be two things: if you want to address a protest you can use the model to find who in the future might join the protest and it can also be used to find how many people will join the protest to arrange additional necessary security measures. It can be used to estimate the general numbers in a protest.
This is more for online, with regards to people going on the road to protest, that we can’t predict using our data. We don’t know who actually goes to protest. It’s more about who protests online.
BTR: In the end, should people be concerned? The 70 percent accuracy and the study in general–is this something we need to be concerned about in the context of government surveillance?
SR: It’s just a paper. This is not being submitted to the government. Our funding didn’t ask for anything, any goal to implement, there’s nothing like this. We have a system called Tweet Tracker, which just tracks tweets. They hardly asked to see it. It’s just to show to the government that this research has been done.
BTR: Why do you think that they wanted the research to be done in the first place?
SR: The paper being written was our decision. I found the material interesting. As long as it was related to the broader area of crisis and analyzing social media during crisis, it’s fine.
Another thing is that everything is public, you know? All the messages [tweets] we looked at were public; there was nothing private that we were going into. If someone wants to protest on social media without everyone knowing about it, they can just make it private. We are not going to hack it.
A lot of being are fine with it being public. They are fine with letting their opinion be known.
BTR: Plans to take the research further?
SR: Theoretically it would be good to link it to offline things but I’m not sure if we can do that right now. Like how people react to crisis and in what numbers.
When something is related to ISIS (or anything radicalization attempt)–intent to do something is totally different from expressing opinion. Expressing opinion is very different than intent to take action on it. We’re trying to bridge the two but it’s a very hard thing to do. So nothing concrete yet.