Hacker News Reader

Comment by convery | Hacker News Reader

We all know that services like this relies on a tonne of statistics, we also know that very little to none is personally identifiable. So it feels like trying to manufacture outrage over it, that it's flaunting how much 'scary data' you have on the users to say that a file has been accessed 56 times when everyones grandma would be fine with a viewcounter on every video.

View on HN

>it's flaunting how much 'scary data' you have on the users to say that a file has been accessed 56 times when everyones grandma would be fine with a viewcounter on every video.

I think it's a different matter. A viewcounter is seen as a stateless counter, increasing for every view/user but not going beyond that; Netflix's analysis is seen as stateful tracking, one that tracks a user through his consumption of data. Also, the former is content-focused, whereas the latter is arguably user-focused.

Even when data is "anonymized", it's sometimes possible to recover identifying information anyway. For example, knowing a person's age, zip code, and gender is often enough to identify people, even if their name is not explicitly part of the available data.

That's just a really simple example. I'm sure you could do much better with advanced AI techniques and more "anonymized" data.

In fact, this was demonstrated on Netflix's own dataset: https://arxiv.org/abs/cs/0610105

I think one of the best and chilling example is still: https://en.wikipedia.org/wiki/AOL_search_data_leak

Just some anonymous search strings can give you people. And that was in 2006 so nowadays with the quantity of informations people make public you'd get a lot more results.

In college (years ago...mid 2010s) I did a panel data regression analysis about certain types of behavior in the dorms. I conducted an anonymized survey, that asked a bunch of demographic and behavioral questions as well as ones about extra curriculars across several years. I was working on a “willingness to take risk” study. When cleaning the data (I allowed too much human entry in the first iteration of the survey) I could read the demo data and see the other activity responses and make quite confident inferences about who some of the respondents were. This Netflix thing is no surprise to me, unfortunately. I sort of feel like there should be some kind of code of ethics for Data Scientists, or maybe anyone that uses econometrics for making inferences about human behavior.

How is netflix's data not personally identifiable? They are tracking your watch pattern to suggest you new viewings. So their stat is connected to the user even if it's not accessible by most employees but it might be. So someone could have written an anonymized query to run this or a query with the names than the final output just containing the number.

Ideally netflix is for the most providing employees/systems with anonymized analysis. Definitely there's always a question of what kind of direct access employees have of sensitive data and I will agree that is a general question even without this tweet.

If Netflix offered a choice[1]: 1) Be part of our data collection program or, 2) Cancel your account, how many people would still pick 2?

[1] They already do this with the privacy policy, right? https://help.netflix.com/legal/privacy

Unless this number is real and you are one of the 53 persons. I would definitely not be ok with the tweet if I was one of them.

My guess it is most likely some store/shop looping the same film(s) and at the same time breaking the contract as public showing of content from Netflix is not allowed.

Call me cynical, but isn't it far more likely that whoever was running Netflix's twitter account that day made up the statistic trying to be funny?

It seems like a weird thing to purposely lookup, however I could imagine someone came across it, "This person watched A Christmas Prince every day for the past two weeks? That can't be right...", and then dug into it more, and mentioned it at the water cooler.

I mean, obviously that's not impossible. It just strikes me as a lot less plausible than "a social media manager thought it might get a lot of retweets".

Personally, if Joe Netflix Employee really can browse viewing data (and I don't know why everyone is assuming that's the case), I'd think stories like this probably wouldn't get shared at the water cooler, just because there'd be so much bizarre behavior that nothing would surprise anyone.

Given the banter I’ve seen on some corporate Twitters (Wendy’s, Burger King) I wouldn’t be surprised if this was fake or only marginally true.

I agree, Netflix to me seems like the strangest website to make a "be careful about your data" point, and doubly so when that particular statistic (53 people watching a specific movie 18 days in a row) doesn't require any identifying information.

thing is, we have no way of knowing companies actually make data not personally identifiable. We need to trust their word. I have heard many stories of data science researchers having detect access to user data...

Reply on news.ycombinator.com