Like-Fueled Algorithm Knows Facebook Users Better Than Their Family

But not quite as well as spouses
 Facebook headquarters entrance sign
LPS.1, via Wikimedia Commons (CC0 1.0)

Every day, Facebook learns more and more about each person who uses it. Clicking a “like” affirms an interest and teaches the machine valuable information about its users. A new study from researchers at Cambridge and Stanford reveals that a few likes are enough for Facebook to score a person on a personality quiz better than actual humans who know that person in real life. With 10 likes, Facebook knows the user better than a work colleague, and with enough likes, Facebook can even out-evaluate family members.

For the study, 86,220 users on Facebook completed a 100-question personality quiz and provided the researchers access to their likes. Using data from 90 percent of the quiz-takers, the researchers matched the likes to traits determined by the quiz. From that, the researchers created a model that uses only likes to predict quiz answers, and applied that model to the remaining 10 percent of quiz results.

To test that machine assessment against human predictors, the people who took the quiz were invited to ask friends and family members to express “their judgement of a subject’s personality using a 10-item questionnaire.” Here’s how well the machine did at evaluating personalities compared to people:

Correlation To Self-Ratings

For every level of friendship — except spouse — the like-fueled algorithm did better than the real-life acquaintance.

This project is hardly the first to mine Facebook data for unique identifying traits or hidden information. This specific study builds off an earlier 2013 study by the University of Cambridge that used likes to identify demographics, even unexpected ones, such as people who like “That Spider is More Scared Than U” are more like to be non-smokers.

The study’s authors see a positive application of the technology, noting that data-driven decisions can improve people’s lives. And what’s not to like about that?