They had a $1 million prize for improving the recommendation algorithm. The data was anonymized and sanitized but it was apparently potentially possible to identify people from some of the data. A closeted-lesbian mother filed a class action lawsuit against Netflix (even though I do not think she was actually outed by the dataset, meaning I do not think anyone ever connected her rental history to her) on the basis that offering the dataset to researchers at all was a violation of US privacy laws over video rental records. The argument was that someone looking at her video rental records could deduce that she was actually gay.
It might be surprising given how lax the US seems to be about privacy in general, but video rental records are protected differently from some other data because of the failed Supreme Court nominee Robert Bork (failed in part because reporters were able to get his video rental records whichI think show he had rented porn, but I don't remember all the details). As a result of the Bork situation, federal law was passed to protect video rental records. Facebook had to settle a thing with the FTC over something similar with its Beacon program.
Netflix wound up canceling the follow-up contest because of the class-action lawsuit. IIRC, a few years later they wound up settling a different lawsuit also related to the FTC stuff, but that wasn't tied to the algorithm or potentially "outing" anyone.
Bork's VHS rentals were actually pretty boring and didn't include pornography. The indignation / backlash to the list's publication made it clear that he and his supporters were hypocrites; Bork had repeatedly argued and ruled in a way to curtail privacy rights.
That lawsuit killed the contest but the recommendation system went on for a while longer until they realized their streaming service didn't have enough content for the original recommendation algorithm to work and they wanted to promote their own stuff.
Clarence Thomas has a way, way more sordid "fetish" history than just renting porn. No normal person would care if it was just him renting an erotic video.
I guess they could have, but data gets stale after a while.
I have to think that the real problem is that the things Netflix wanted to start to optimize for different things rather than recommendation accuracy. They have new metrics like watch time, time engaged with the app (binge watching), watch to completion, rewatch time, watch first weekend, etc. I don't think they care about offering the best recommendations anymore, just about doing whatever they can to keep you tuned into the app as long as possible. Those two things aren't necessarily at odds, but they aren't the same thing either.
Like, today Netflix is going to want to optimize for getting the most eyeballs on its originals versus content they've merely licensed from the catalog. When it was offering DVD rentals, it didn't have original content so having good recommendations was the sort of thing that would keep the service "sticky" and prevent subscriber churn.
> having good recommendations was the sort of thing that would keep the service "sticky" and prevent subscriber churn
It's the opposite. The primary cause of churn is that there is nothing a subscriber wants to watch. You might assume that a recommendation helps a user maximise the amount of the library they will watch but it actually helps the user conclude there is nothing they want to watch and leave.
Netflix wants their library to be more like a one-armed bandit. Users have to gamble on whether they will like something and they will never know they have run out. They want the user to keep hanging on, to keep scrolling, assuming something great is just around the corner, never getting to "I'm done with this".
I noticed this effect with the DVD subscription. If you added 3 discs to your queue, they would send the first two but not the third, waiting for you to add some less desirable titles to your queue, which they would send before they sent your high priority title. If your needs get satisfied, your subscription ends. It's a fundamentally user hostile business.
Netflix' key metrics are basically around retaining subscribers and attracting new ones. Except to the degree it supports those goals (and there's obviously a correlation), they don't actually care how much you watch, how much you like the content, how easily you find content you like, etc. Correlated, yes. But not one and the same.
If you open Google and type in “netflix recommendation system lawsuit” you’ll find many results on the first page explaining it. You can use this method for a lot of topics that people here on hacker news mentions.
Could you please stop posting unsubstantive comments and flamebait? You've unfortunately been doing it repeatedly. It's not what this site is for, and destroys what it is for.
Your low effort comment is a hundred times worse than the one you responded to. Trying to shame someone in that sarcastic juvenile tone for asking a question in good faith is itself shameful.
People naturally respond to such things and use them to segue into another topic as part of conversation, and sometimes it is nice to have someone tell their recollection of a story instead of getting the wikipedia version.
It might be surprising given how lax the US seems to be about privacy in general, but video rental records are protected differently from some other data because of the failed Supreme Court nominee Robert Bork (failed in part because reporters were able to get his video rental records whichI think show he had rented porn, but I don't remember all the details). As a result of the Bork situation, federal law was passed to protect video rental records. Facebook had to settle a thing with the FTC over something similar with its Beacon program.
Netflix wound up canceling the follow-up contest because of the class-action lawsuit. IIRC, a few years later they wound up settling a different lawsuit also related to the FTC stuff, but that wasn't tied to the algorithm or potentially "outing" anyone.
[1]: https://www.wired.com/2009/12/netflix-privacy-lawsuit/