Your Marital Status Is Hidden In Your Tweets, Algorithm Reveals


In the future, online lovers may be able to fire up an algorithm to check their crush’s relationship status instead of just stalking their Instagram. A team of mathematicians recently announced they’ve created an algorithm that incorporates data from multiple social media accounts to predict whether a person is married or single. So far, the system has been highly successful in its analysis of most individuals, save for one person in particular — the president of the United States, Donald Trump.

At the AAI Conference on Artificial Intelligence in February, the Chinese and Russian researchers revealed that this algorithm was able to predict whether or not a person was single or married with 86 percent precision. This, they further explained in a new but unpublished paper, is 17 percentage points more accurate than previous, similar models. The data they used to build and test their algorithm were taken from the Twitter, Instagram, and Foursquare feeds of thousands of people from Singapore, New York, and London. The researchers gathered data on average tweet size and number of check-ins, plugged it into a machine learning model, and then compared their algorithm’s analysis to survey data to determine whether the user was actually single. More often than not, the algorithm was right.

But its limits were revealed when the algorithm got Trump’s marital status wrong. After it was fed only tweets from Trump and Barack Obama, the algorithm suggested that while Obama was married, Trump is a bachelor.

This discrepancy, the researchers explain, is rooted in the fact that Trump doesn’t always write his own tweets. “We all know about his wife Melania,” said study co-author and ITMO University associate professor of computer science Andrey Filchenkov in a statement. “But in this case, we are studying whether all Trump’s assistants are married or not. We are not guessing who Trump is, but who runs his social media.”

Actually, we often do know who is tweeting from the Trump account. Several media outlets have pointed out that the President’s own tweets are the ones that are tweeted from an Android phone, leaving us to presume that his staff’s tweets are the ones sent by iPhone. When it comes to Obama’s tweets, however, the general public doesn’t know if they are composed by him or his staff. This inconsistency makes it easy for the algorithm to produce inaccurate results.

It also underlies the question of whether or not someone’s data can really reveal the truth about their character. When the data is a fair representation of the user, it appears to be so: At 86 percent, this algorithm’s accuracy rate is impressive. Similarly, a model designed in 2011 was able to accurately predict an individual’s score on a personality test by simply scanning the words they use on Facebook.

The researchers behind this new algorithm hope that it eventually can be used as a way to create a personality profile based on a person’s public profiles, which will be useful to job recruiters who wish to get a better snapshot of who they’re hiring. But it doesn’t seem improbable that this sort of program could be used by people trying to catch catfish as well.