Past research has found that by looking at what you’ve liked on Facebook, machine learning algorithms can predict your personality traits better than your own friends, family, and spouse can.
Last week, I covered a recent paper by Michal Kosinski on predicting politics from social media photos, which is a result that doesn’t have a clear explanation. Is there really something about a person’s face that reflects what political ideas they will like? Using Facebook likes to predict personality, on the other hand, is much more intuitively plausible, and there are potential mechanisms to examine. Understanding the implications of this finding, however, requires understanding more about the way studies like this are set up.
When psychologists talk about personality, they are generally referring to the Five-Factor Model of personality, which captures broad-based aspects of personality. The five factors of Extraversion, Agreeableness, Conscientiousness, Neuroticism, and Openness to Experience come from looking at the way people rate themselves on every possible descriptor in the English language. These five factors represent major patterns in the way people describe themselves. For example, people who rate themselves as highly outgoing also rate themselves as highly talkative and highly assertive. Statistical tools are used to identify common patterns in responses and create the dimensions. For example, the descriptors ‘outgoing’, ‘talkative’, and ‘assertive’ would all fall under the common dimension ‘extraversion’.
The machine-learning algorithm that predicts personality from Facebook likes is therefore capturing ‘big picture’ aspects of your personality, and not nuanced details like whether you are outgoing at parties but keep to your group at bars. Further, the algorithm is capturing the way you see yourself, because personality scores are based on your own ratings of how much you agree with statements like “I am someone who is outgoing, sociable”.
The reason Facebook likes can plausibly give insight into personality is that they reflect cultural products — bands, sports teams, products — that have specific images, often crafted to appeal to certain types of people. If I told you that someone likes Cardi B, you might have an idea of what that person is like — maybe that person is a little less neurotic than the average person. If I also told you this person likes Sephora, you might guess the person is a bit more extraverted. If I gave you enough information about the various things a person likes then eventually you would be able to guess, broadly, how that person sees themselves. This is what the machine learning algorithm is doing.
There is a very interesting graph in the paper that shows how having information about more things a person likes leads to better and better accuracy in personality prediction. If a person has liked 10 things on Facebook, the algorithm can guess the person’s personality about as well as a work colleague can. If a person has liked around 70 things on Facebook, the algorithm can guess personality about as well as a friend or roommate. At just under 150 likes, the algorithm can predict your personality as well as a family member can. By the time it’s seen 300 different things you like, the algorithm can actually beat your spouse in predicting your personality.
It may seem surprising that an algorithm can guess your personality better than your spouse. But it’s worth mentioning how accurate that guess is: 34 per cent. Across all the Big Five traits, a spouse is typically only 34 per cent accurate, and that’s also approximately how accurate the Facebook likes model is. It turns out people aren’t great at guessing how others see themselves on broad personality traits. Considered in this light, it may not be so surprising that if I gave you a list of 300 products, bands, politicians, and other cultural products someone liked, you could get to about ⅓ accuracy in guessing their personality.
It’s potentially more interesting to consider what the predictions mean, and how they are made. First, you would need to know what liking each of the various Facebook pages means in terms of personality. The only way to reliably learn those relationships is to have a huge training dataset, where you could observe that, for example, liking Jonathan Van Ness is reliably associated with being more agreeable.
Second, you would need to know whether these associations will be consistent for a while, or whether cultural tastes will change. For example, liking the brand Nike might have been associated with one personality profile before they chose Colin Kaepernick as a spokesman, but been associated with another personality profile after that decision.
Third, it would be interesting to consider whether certain patterns of Facebook likes have distinct meanings. For example, does being a fan of the K-Pop groups BTS, SuperM, and BIGBANG all mean the same thing? Or does liking different clusters of K-Pop groups tell us something more specific about the person’s personality?
This is where the careful application of the scientific method can combine with machine learning algorithms to provide deeper insight. Knowing that you can predict personality accurately from what cultural products a person says they like is interesting, and machine learning algorithms are useful for finding robust and complex relationships. What is more interesting, however, is digging into the details of these relationships to come to a deeper understanding of how this process works. — Psychology Today
Alexander Danver is a Postdoctoral Fellow at University of Arizona
He is the best to lead the country to a new era of growth
Opinion1 week ago
Country is a melting pot of not just cultures, nationalities and religions, but also diverse enterprises and business models
Opinion1 week ago
Tesla chief's comment on Japan's population is an irresponsible rush of going down the rabbit hole
Opinion1 week ago
Although India’s poorest half holds a small percentage of the country’s total wealth, its members are still better off than their peers in most countries
Opinion2 weeks ago
For the longest time in America (and indeed in much of the world) women were underrepresented in most of the professional workforce
Opinion2 weeks ago