Hacker News Reader

How would that work for site like Facebook or Instagram or other sites where there is no clear relevant content to target?

View on HN

There's been a lot of work in this area over the last few years, clustering and tagging text and video. Facebook even has an open source library for it called Starspace.

But the point is that Facebook is showing you ads based on your surfing behaviour outside Facebook - not the content you view on Facebook.

If you use facebook purely for posting and viewing family baby photos, but view websites about tractors, you'll probably see ads on facebook from tractor related companies.

Starspace is used by FB to tag the sites that report that you've visited them to facebook. So they can target ads

I think the point is that FB could change to a contextual model. E.g. after seeing a post of a news article by Tractor Daily you would see an ad for a new John Deere.

Can you be more precise to what you means? That sounds like targetting.

Natural language processing for posts and video analysis via machine learning could work.

Reply on news.ycombinator.com