Facebook ‘very poor’ at distinguishing political ads, KU Leuven researchers find

Facebook ‘very poor’ at distinguishing political ads, KU Leuven researchers find
Photo by Solen Feyissa on Unsplash

Facebook misjudges up to 83% of the advertisements deemed political, according to new research from KU Leuven and New York University.

In some cases, Facebook doesn’t recognize political advertisements for what they are; other times they wrongfully label non-political content as political.

“This ultimately misleads both users and advertisers on Facebook: either it is unclear that an ad is actually conveying a political meaning, or ads are wrongly removed because Facebook believes they are political,” Victor Le Pochat, PhD researcher at KU Leuven, told The Brussels Times.

“The harm this causes is twofold – it can be economic because you're preventing normal advertisers from publishing commercial messages, or even taking away valuable public health messages regarding the pandemic from the public; it also causes harm by providing an opportunity for advertisers to circumvent the rules, opening the door to misinformation without having to reveal who they truly are.”

The researchers say that these shortcomings also erode public confidence and undermine trust. This contradicts the entire reason for labelling political advertisements which was intended as a means of building public confidence following the 2016 US election and the Brexit referendum.

Disparities between countries

The research reflects Facebook’s work on a global level but also highlights differences between individual countries: Facebook's labelling was worst in Malaysia with 45% of political advertisements not getting the proper designation.

However, the labelling policy worked better in the US and New Zealand, where only 1% of political advertisements there were missed. But for the US, this still amounts to just under 10,000 advertisements given the high quantity of political advertising that takes place on the platform.

“Facebook could take some simple measures to improve its detection of political ads, but it has already indicated that it doesn't feel particularly inclined to do so,” Le Pochat said, explaining that the reason it performs better in the US and New Zealand is likely because it’s optimised for the English language.

“It affects the European community disproportionately because we have a lot of countries that don’t speak English and a lot of countries with smaller communities. They miss a lot more ads in countries in Europe. They don't really invest in making their systems work well outside the US.”

Facebook has required political ads to be labelled in order to indicate who is paying for them, but the onus falls largely on the advertisers themselves to label their ads and not all of them comply.

An imperfect algorithm

An algorithm is supposed to catch any advertisements that were not properly declared, but the study proves its success varies highly.

A list of recommendations from the researchers includes imposing consequences for the advertisers who fail to properly categorise their advertisements and engaging with local governments, regulators and organisations in order to better understand the local context.

Le Pochat thinks that there’s not only plenty of room for improvement when it comes to political advertising on social media in Europe, but also greater will and more ways given the existing EU legislation on data privacy and advertising transparency.

“We’d like to see mandated transparency for all ads, not just political ones. We’re calling for universal ad transparency with enough detail made public for us to assess who’s behind it and who’s paying for it, then hold those people accountable,” said Le Pochat.

“But these measures should be sufficiently comprehensive and complete, otherwise we have a partial solution and we’re not taking steps forward – we may even be taking steps backwards.”

Copyright © 2024 The Brussels Times. All Rights Reserved.