[spunker540]

Photos and videos of people are not generally useful for ads. You know what is valuable? Actual purchase decisions sent in by advertisers. Photos and videos of people’s dogs and food and kids etc is strictly to entertain the other Facebook users and keep them coming back.

An advertiser is willing to pay tons more to serve an ad to someone who has bought similar products in the past, and is likely to buy their product again, or has recently abandoned a non-empty shopping cart. Or has even just interacted with a previous ad. That is the data that increases ad revenue. Not facial expressions while scrolling or group photos at restaurants.

[mceachen]

I'd hesitate to discount the value of photos.

Photos have quite-accurate GPS and frequently-accurate compass heading. Add face detection and you have location tracking for everyone in frame.

Online tracking is valuable for ad targeting, but physical tracking is as well, especially for brick-and-mortar businesses.

[danShumway]

You're saying this with a lot of confidence, but I'm not certain I believe you.

When you have facial recognition software, photos are a contact graph. When they're photos that a user explicitly uploads, they often come with location data. All of that stuff is useful to advertisers.

Facebook can and does target ads based on who you know. If you upload a photo, and the metadata says it was taken today, and it has you location attached to it, and it shows you smiling next to your friend who's recently searched for the new Avengers movie, then yeah, it makes a lot of sense to show you an add for the new Avengers movie, because maybe the two of you will go to it.

It doesn't even need to be that detailed. Just knowing that you're that person's acquaintance means that now Facebook can suggest that person as a friend, it means that Facebook can add that person to its social graph if they're not currently on the service. Just knowing your location means that Facebook can advertise you something from a nearby store.

All of that is stuff that Facebook wants to know, all of that falls pretty squarely in the middle of the category of data harvesting. Not all data harvesting is just to serve ads, some of it is to know how to tailor the service to you to make you more active (so that you look at more ads), or to suggest friends (which increases your investment in the platform), or how to engineer the service in general (so that more people use it and look at ads).

> Not facial expressions while scrolling or group photos at restaurants.

Facebook used to analyze unposted status updates (where you type out the status and then delete it before you press post) so it could figure out why people were getting cold feet before sharing their thoughts[0]. You really, genuinely believe that they're not interested in your facial expressions? You really, genuinely believe that nobody at Facebook is looking into research like identifying branded products inside of user-uploaded photos?

[0]: https://slate.com/technology/2013/12/facebook-self-censorshi...

[reaperducer]

Photos and videos of people are not generally useful for ads

If that's true, then how come there was a big hoo-ha a few years ago about Facebook putting pictures of people's friends in ads for unrelated companies, as if the friends were endorsing those products?

Sounds like a very useful way to use someone's photos for ads.

Also, with the state of machine learning, it's hard to imagine how image processing wouldn't be useful for advertising.

I can type "cat" into my phone, and its on-device learning shows me all the pictures with cats in them.

Facebook has the technical ability to scan people's pictures and tell all the cat food advertisers, "Here's all the people interested in cats." Or, even more precisely, scan the photos products or logos and tell an advertiser, "Here's all the people who use your competitor's product."

It's not hard to see where this is going, assuming it's not already there.

[danudey]

Photos and videos of someone and where they are is valuable for ads. The logo on the beer you're holding in your photo with your friends at the bar, the faces of those friends and what drinks they have, the geolocation information in the image metadata, the date and time the photo was taken, and so on. All of that is valuable data to Facebook, and they're operating at a scale that allows them to fully harvest and take advantage of all of that and more.

At Facebook's scale, even tiny, inconsequential data can be useful in aggregate.

[soylentcola]

I would imagine that on a large enough scale, analysis of photo contents, timestamp, location, and correlation with any other FB/app activity at that given time/location can provide a ton of new data points to add to their behavior modeling algorithms.

There may not be obvious one-to-one connections that we would make from the information we see in a photo, but the more data points that can be fed into a model, the more accurate the model becomes.

[heavyset_go]

> Photos and videos of people are not generally useful for ads.

They're great for facial recognition, finding out more about relationships that exist in real life, and building sets of training data. Facebook is in all of those markets.