AdNews has dipped its toe into the pool of ‘interests’ that Facebook assigns users for advertising purposes and of the sample survey we looked at, the social media giant got user ‘interests’ right 40% of the time.
The findings here are based on a sample of 10 users – a small fraction of Facebook’s 2.1 billion user base – and do not reflect the accuracy of its broader targeting capability, which also uses self-reported information about a user’s geographic location, gender, age, political preference and other demographics, to help advertisers reach an audience.
When looking at interests, given some are hit and miss, it raises questions about current machine learning technology and how accurate it can be when it comes to predicting things such as a person's favourite sports teams, food, fashion or hobbies.
Based on this small sample, the social media platform is most accurate in working out people’s interests in fashion, food and relationships, but scored poorly in fitness, education and hobbies.
‘Interests’ are one of the ways advertisers are able to target relevant ads to consumers on the social media platform, and evolve the more the platform is used.
Facebook told AdNews that ‘interest targeting’ reaches people based on the apps they use, the Facebook pages they liked and the ads they click on and off the Facebook platform.
Facebook also takes into account demographics, such as age and gender, in predicting a user’s interests.
“Our platform aims to connect people with relevant advertisers at scale. The more relevant an ad is, the more likely someone is to find it useful and engage with it,” a Facebook spokesperson told AdNews.
The findings
AdNews tested out interest accuracy by asking 10 people to note all of their Facebook interests in the 14 different categories, which includes things like hobbies, news and entertainment, sports, travel, fashion, lifestyle, food and more (see infographic).
The sample included six females and four males from different occupations.
Participants were asked to mark each interest either ‘1’ for relevant, ‘0.5’ for partially relevant or ‘0’ for not relevant at all.
The sample provides an indication of how much interest accuracy varies between users, but is too small to provide any conclusive evidence of what the real accuracy score would be across the vast social media network. These results should be interpreted with this in mind.
Facebook assigned a combined 3573 interests to the 10 users, which equates to about 357 interests per person.
After marking for accuracy, the sample’s combined interests score was 1343, which averages out to 134 accurate interests per person.
The average accuracy rate was 39.4%, while the median accuracy was 40.8%.
Accuracy scores varied for each person, with the highest at 60.5% and the lowest at 8%. Five participants had an accuracy score of 45% and above.
By category, Facebook was most accurate in shopping and fashion (getting 52.7% of interests right), followed by food and drink (52.1%), family and relationships (50.9%), technology (46.8%) and travel, places and events (45.2%).
The least accurate categories for this group were education (22.7%), fitness and wellbeing (25.2%) and hobbies and activities (28.1%).
The range of accuracy across categories also fluctuated significantly across group, particularly as most participants had one category that had zero accuracy.
The highest range was 91.3 percentage points and the lowest was 29.17 points, with a median of 65.4%.
Why this matters
Targeting has become increasingly important to advertisers trying to reduce wastage and improve the cut-through and effectiveness of campaigns by providing users with relevant ads.
This is particularly the case with digital display advertising, which has been tarnished by years of intrusive and annoying ad units on websites fuelling the rise of adblockers.
Facebook and Google are widely regarded as the most sophisticated digital platforms for targeting.
In the case of Facebook, it’s important to note that it allows users to monitor their ad ‘interests’ and delete the ones that aren’t relevant, providing users with transparency and control over what ads they are served. A user’s interests also naturally evolve the more a user interacts with the platform.
AdNews asked Facebook how accurate the interests were and was referred to a 2016 blog by VP of ads and business platform, Andrew Bosworth.
“While more than a hundred companies already serve interest-based advertising on websites and apps today, we offer a better experience because we care about the integrity of Facebook ads,” Bosworth wrote.
“Ads are reviewed against our standards and to ensure they are as respectful of people’s experience as possible.
“For example, we don’t permit ads that include sound unless you interact with them and we prohibit deceptive ads and ads for unsafe products and services. We’ve developed technology to determine when someone clicks on an ad on a mobile device by accident, so you don’t get taken to a website or app you didn’t mean to visit.”
Facebook openly admits its interest-targeting technology is not perfect and it doesn’t always get it right.
This study sheds light on the challenges media companies face in trying to accurately predict what a consumer likes even with one of the largest pools of consumer data on the planet.
Have something to say on this? Share your views in the comments section below. Or if you have a news story or tip-off, drop us a line at adnews@yaffa.com.au
Sign up to the AdNews newsletter, like us on Facebook or follow us on Twitter for breaking stories and campaigns throughout the day.