Slashdot Mirror
Using Facebook Data, Algorithm Predicts Personality Better Than Friends
sciencehabit writes: A new study of Facebook data shows that machines are now better at sussing out our true personalities than our friends. One of the standard methods for assessing personality is to analyze people's answers to a 100-item questionnaire with a statistical technique called factor analysis. There are five main factors that divide people by personality—openness, conscientiousness, extraversion, agreeableness, and neuroticism—which is why personality researchers call this test the Big Five. People can accurately predict how their friends will answer the Big Five questions. ... Compared with humans predicting their friends' personalities by filling out the Big Five questionnaire, the computer's prediction based on Facebook likes was almost 15% more accurate on average, the team reports online today in PNAS (abstract). Only people's spouses were better than the computer at judging personality.
The comment that the algorithm does better at predicting personality than a person's friends will depend very strongly on how you define a friend. I have a very large number of Facebook friends about whom I know almost nothing, so I am not at all surprised that an algorithm will do better.
I am a Statistician. One false move and you are a Statistic
And yet, I got dirty looks in church on Sunday because I didn't know somebody was seriously ill for a month with pneumonia. Apparently, everybody (but me) has been talking about it on Facebook and if I don't know I'm the bad guy.
Peter predicted that you would "deliberately forget" creation 2000 years ago...
Why are people still using Facebook? Because other people are, and they use it as their medium to schedule events and coordinate activities.
90% of my Facebook activity is devoted to participation in a handful of secret/private groups, and the other 10% is responding to event invites -- some of which are "go, no-go," others are FCFS based on responses to the invites.
Also, I mostly DNGAF about Facebook (or Google, or whomever) knowing what flavor potato chip I prefer because I used my club card at the store. Google gave me $15.98 on their Opinion Rewards platform for knowing even MORE about me. Whee!
The names of the factors are guesses. Factor analysis looks at the covariance matrix of items, and finds sub-matrices of the total matrix that meaningfully covary. Each one of those sub-matrices is called a factor, or latent variable, which is measured by common covariation between the questions. The number of latent factors found in a questionnaire is typically derived both by theory (we made a questionnaire intended to measure these 6 different things) and empirical facts (of which typically would be Horn's parallel analysis or the Kaiser criterion [which simply means all eigen values of the covariance matrix that are greater than one]). The factors are named because that is what was a suitable commonality between the items first measured, along with external criterions like predicting other theoretically related constructs. The Big 5 are an enormously well studied problem space, and the stability and pervasiveness of these concepts have been well documented and linked to specific gene expressions, developmental trends, et cet.
It doesn't matter if you use Facebook or not - they can already infer that TV shows and musicians exist via user data and automatically construct pages for them - WKRP In Cincinnati is a good example or was when I looked at it last summer. If they can infer media exists then it stands to reason that they can infer that you exist. Imagine that, if you will - a near future in which you have a fairly accurate social media profile rather you want one or not.
Haven't you failed to read the article before claiming that it is wrong?
For those playing along at home, Fig.1 from the actual article explicitly refutes the AC's claim.