A team of researchers from Facebook has come up with a way to predict the ethnicity of Facebook users in the U.S. by using a combination of U.S. Census data and an analysis of users’ first and last names. The team won the award for best paper at a social-media conference organized by the Association for the Advancement of Artificial Intelligence in Washington, D.C.
The Facebook Data team confined its analysis to the country’s four largest ethnic groups: Caucasians, African-Americans, Asian/Pacific Islanders and Hispanics. It used Census records to identify the most likely ethnicity of a user based on his or her last name. For example, said data scientist Jonathan Chang during his presentation, if your last name is Mueller, Census records show you have a 97% chance of being white. Similarly, if your last name is Washington, there is an 89.9% chance you’re black.
The team then refined its predictions by factoring in users’ first names, Chang said. That helped reduce confusion, for example, between Caucasian and African-American users with the same last name. A user with a first name such as LaToya was probably African-American, Chang explained.
By applying their predictions of ethnicity to users’ friend networks, the team found:
- The ethnic makeup of Facebook users has steadily become more diverse and now generally reflects the U.S. population, unlike a few years ago, when Caucasians and Asian/Pacific Islanders were over-represented.
- Users are more likely to be friends with, and communicate most often with, people of the same ethnicity.
- Users are more likely to be in romantic relationships with people of the same ethnicity.
- Particular ethnic groups tend to behave similarly online, with Asian/Pacific Islanders engaging in “unexpectedly high number of wall, video, note, gift, comment and group-sharing actions.”
The team acknowledges a few caveats: Its analysis didn’t account for a lot of factors, such as socio-economic status and education, and it didn’t include smaller ethnic groups in the country. The researchers plan to use more detailed Census data to factor in user locations, professions and other self-disclosed data to “improve the predictive power” of future analyses, the paper concludes.
So what does this mean for businesses looking to refine their marketing on Facebook and other social networks?
Approached after his presentation, Chang said he knew little about business and could not answer how his team’s findings would apply to consumer marketing.
But surely, now that it’s possible to predict ethnicity on Facebook with relative accuracy (accuracy that will only improve with subsequent studies), the next logical step would be to allow marketers to target users based on ethnicity and not just age, location and known preferences.
It would be a simple matter to rotate racially appropriate photos in Facebook ads aimed at 30-year-old women in the Washington, D.C., area, for example, in the hope that they would respond better to someone from their own race.
What do you think — does that kind of marketing cross a line or is it just one more refinement on what advertisers do all the time?
Image credit, Rich Hobson, iStock Photo