Hacker News Reader

Comment by dymk | Hacker News Reader

Age and gender are the two highest signals that targeted ads take into account. The Facebook growth team has always wanted to remove gathering age / gender from the sign-up flow (which reduces friction to sign up). But, ad targeting would take such a large hit, they could never do so.

It was a huge deal internally when "Custom" was added as a gender option in the sign-up flow, and even then, there's a nudge towards the binary to retain better ad targeting.

In addition to that, companies wants you to acknowledge that you're over 13 so they can be more loosey-goosey with data collection (thanks California!).

View on HN

> It was a huge deal internally when "Custom" was added as a gender option in the sign-up flow, and even then, there's a nudge towards the binary to retain better ad targeting.

It's odd that advertisers would dislike that change for targeting -- surely the people who pick "custom" are sending a major signal that correlates with other preferences. It's a third grouping they can specifically target! If they split "male" into "HYPER-MASCULINE ALPHA MALE: CARS, GUNS, AND FOOTBALL FUCK YEAH!" and "I have a penis but I try not to let it make decisions for me", advertisers should presumably love that distinction too. Of course the reality is that their decades of research deciding what constitutes "male" and "female" has gotten them stuck in a shitty local minimum and they can no longer abandon this binary concept that they cling desperately to without admitting that they've never actually known what the fuck they're doing.

If there was a 300 dimensional vector expressing your nuanced gender identity that would be even better, but advertisers would have no idea how to target it (at least at first).

> advertisers would have no idea how to target it (at least at first).

It would take very little time to reap massive efficiencies in targeting, and the rest would be long-tail that would be discovered as budget allowed.

The easiest way to do this would be to train a bandit model to be a bidder, using this vector as part of the input features.

You would spend some time and money doing randomized exploration (maybe something more sophisticated with active learning, but just uniform random could work), then train a basic bidder to optimize whatever you are currently interested in.

The process of manually deciding which tags fit which ads is not long for this world.

> It was a huge deal internally when "Custom" was added as a gender option

Did a significant amount of users pick this though?

Why in the world do they then collect all that data on us in general if ads still use age and gender for the highest value signals?

Those are terrible signals and really should be replaced with something better.

They're not terrible signals. They're by far the highest-signal metrics that have been found for ad targeting. Legions of engineers and data scientist and project managers have been working on finding better metrics for a decade, and there hasn't been one found.

Why are you paying people who can't figure out my age and gender from my behavior?

Funny if 50 odd genders that Facebook allowed you to select internally mapped to two.

More like four: Maleness true/false, and femaleness true/false.

But gender is a social construct? Why should such a meaningless data point have any impact on their bottom line?

Money is also a social construct, yet I suspect it has a decent impact on their bottom line.

How do you mean? The thing you are measuring affects your measurement?

Your assumption is the data is meaningless, yet the data shows that excluding self-reported gender from ad targeting leads to worse ad performance...

I mean it literally. Money is a social construct. The joke here being that it's funny/ironic to claim that gender being a social construct is "meaningless" but then turn around and claim that another social construct -- money (which is what the "bottom line" is made of) -- is meaningful.

The fact that something is a social construct doesn't mean it doesn't matter. Heck, once you get past the basic "keep this biological organism alive" level, most of human life is lived within social constructs. That doesn't meant they're not real or not meaningful.

Gender is a social construct in that various 'gender typical behaviours' don't necessarily correlate with sex between cultures. That implies it is a learnt behaviour.

In a society where someone of the male sex is pressureed to be male gendered there will be a larger correlation between gender and sex. Which would mean miracle balding cures would do better targeting the male gender if they don't have the inofrmation to target the male sex.

Who told you that social constructs are meaningless? They're not. That's not what the term means.

Reply on news.ycombinator.com