MIT study on Implicit relationships

-->

I was reading somebody’s blog post and watched the link they posted about the UN’s Big Data research when I heard the emcee reference a study out of MIT that looked at Facebook user profiles at MIT and could predict a user’s sexual preferences based on their friendships. So, I investigated the research.

 You can read the entire paper here. The title is 'Gaydar: Facebook friendships expose sexual orientation' by Carter Jernigan and Behram F.T. Mistree.

Disclaimer: I didn't read the entire paper. It's really long. I did read the majority of it.

Here are the high points:

Aggregation of personal data –  The researchers talk about how in any environment – state department or social media, the aggregation of data poses a great potential for risk. A few examples they use are officials looking up the passport activity of President Obama as well as confidential information of taxpayers. If the information is available anywhere, people will exploit it. 

Sex segregation – As the old saying goes “birds of a feather flock together” and this truth intuitively makes sense. Those of us that enjoy leadership and being on teams will migrate towards campus involvement. Those more proficient in sports activities and have a competitive attitude and drive and thus will most likely find each other on sports teams. Well, the same could be said in regards to how sexes segregate themselves. A study cited in the paper states that men have 65% male friends and 35% female friends and females have 70% female friends and 30% male friends. This suggests a likelihood that if given an individual person identified by a specific sex, they should have the same percentage of friendships. A significant deviation from this might offer some insight to that person's lifestyle.

Homosexuality and sexual segregation – According to the study, homosexual men and women draw the majority of their friends from the LGB community whereas bisexual men and women draw their friends from the heterosexual community. So, again an observation of the amount of the sex of friendships of an individual user should begin to indicate insights about that user.

Forming the hypothesis – Because of these observations, we expect gay males to have a higher proportion of gay male friends online.

Methodology - The researchers used a web-crawling software (Arachne) to go through all of the Facebook users at MIT and extract data concerning student’s sex and “interested in” information as well as the user's friendships. The result was that the researchers could use explicit and implicit (see article for explanantion) friendships to deduce a user’s sexual orientatation based on what their friends listed in their “interested in” bio information. For example, if an MIT male was examined, based on the percentage of male/female friends he had and their listed sexual orientations – the algorithm developed by the researchers could tell the sexual orientation of the user.

Implications – So, the most interesting thing about this, to me, isn't being able to know someone's sexual orientation.  That's interesting, but I don't really care.  Rather, what's really significant about this is the notion that people can harvest seemingly harmless information about you and use it to make implicit assumptions about you. Think about it this way - you're a CIA agent. You live in Atlanta and work for a private equity firm as a cover so your family and friends think you're normal. Say you're trying to keep up with your kids so you have a very basic Facebook profile and every once in awhile you tweet a picture of you and the fam on vaca or the new boat you just bought.  Obviously, you wouldn't post that you work for the government in your 'About Me' information or anything that would significantly link you to a clandestine profession. But, what this research is suggesting is that we can make accurate predictions on a user's lifestyle/personal habits based on information that they aren't making readily available. So, perhaps I scan your friends on Facebook and most of them make sense except a few that are located in D.C. and that information coupled with some of your spending habits that I've observed on your Twitter/Instagram accounts yields a suggestion that you travel to cities that are outside of your job's requirements and your standard of living is significantly different from what it should be. Certainly there are other explanations for these observations. However, another insight that this research suggests is that "types" of people act in similar patterns. So, sure all of these insights into your buying history and friendships would be normal ordinarily. However, when compared with a test set of other CIA agents' info, we come up with a 95% likelihood that you are, in fact, a CIA agent. Cover blown. Your kids and wife are taken hostage and Arnold Schwarzenegger is called in to come and save you.

This is a bit of a reach, but I think the underlying coversation here is very significant. If Big Data analysis is allowing businesses, governments, or (God forbid) terrorists to gain useful insights into aspects of our lives that we aren't intentionally sharing... what could that mean?