Facebook Published Some Info on How They Use Machine Learning to Deliver Ads

Facebook published some info on how they decide which ads to show people? The two main factors: audience targeting selected by advertisers and the results of our ad auction.

The first point is simple advertisers choose their target audience through our self-service tools. Audiences are created based on categories like age and gender, as well as actions people take on our apps such as liking a Facebook Page or clicking on an ad. Advertisers can also use information they have about their audience, like a list of emails or people who’ve visited their website, to build a custom audience or a lookalike audience.

For ads that enter the auction, Facebook selects the top ads to show to a person based on which ads have the highest total value score — a combination of advertiser value and ad quality.

• Advertiser Bid – How much money you’re spending on your Facebook ads – the more you allocate, the more this will contribute to reach
• Estimated Action Rate – Facebook estimates the likelihood that each user will take action on an ad, based on a range of factors relating to their individual behaviors
• Ad Quality – Facebook measures this based on feedback from users (e.g. how many people report or hide your ad) and “assessments of low-quality attributes” (too much text in the ad’s image, sensationalized language, engagement bait).

They find advertiser value by multiplying an ad’s bid by the estimated action rate. This is an estimate of how likely that particular person is to take the advertiser’s desired action, like visiting the advertiser’s website or installing their app. They then add the ad quality score, which is a determination of the overall quality of an ad. This is where machine learning comes into play.