Netflix's Viewing Data: How We Know Where You Are in House of Cards | Hacker News Reader

Netflix's Viewing Data: How We Know Where You Are in House of Cards

techblog.netflix.com

79 pointstrickz46 comments

I wish they would note which movies they've shown to me and I've chosen not to watch 1 bazillion times. Then stop showing them to me. For recommendations, isn't knowing what I chose not to watch as important as knowing what I've actually watched?

The Netflix UI is not completely consistent, so on some screens the option is not available, but in theory you can rate a movie as "Not Interested" for these cases.

Haven't see a "not interested" option in the iOS app. Where does it show up?

Not sure on iOS. Here it is on the web interface:

https://imgur.com/O1DaihN

Yup. Similarly, I wish they'd stop recommending entire sets of options based on a TV show I abandoned halfway through the first episode three months ago because it was so bad.

Just go to https://www.netflix.com/MoviesYouveSeen and remove said show from your history. I do this every time our guests with small children watch cartoons, but fail to switch into "kids mode".

They also bizarrely consider something that you have only watched a few seconds of (by accident) as watched and like to show recommendations by that.

Through human eyes, Netflix's recommendations are especially dumb for the reasons you and I give. Equally pointless is recommending things I have already watched in a genre instead of other items in that genre, and doing a filter bubble where they keep narrowing down to things like you have already watched.

My favorite example of these "dumb" (to human eyes) recommendations happened to me a year ago: https://twitter.com/chimeracoder/status/423877639666167808. Note the reason Netflix recommended it.

I have no trouble understanding why a recommendation engine might yield this sort of result (e.g. parents sharing accounts with kids), but I'm a bit surprised that Netflix doesn't do a better job of inferring when different people are using the same account, even when the users don't keep their profiles separate explicitly.

The amusing part is that I've never even seen the show myself. My friend happened to watch a single episode of Law & Order: SVU on my computer while she was staying at my place... and now I'm probably on a government list somewhere after this recommendation!

In reality though it's a good recommendation (for your friend, not you), it just seems odd if you have a particular mental model of what a recommendation should be like, which this violates.

As you say the demographic that watches SVU is likely to have kids that watch Barney and may not be aware that it's on Netflix, so this is a good recomendation for parents even if they won't watch it immediately after some late night SVU binge but rather the next day with their kids.

I remember Netflix recommending a whole bunch of random cartons and The Sons of Anarchy for my 4 year old daughter on her profile.

This is a classic example of focusing on fancy, clever things and forgetting about the obvious and simple wins.

I wonder if this is intentional? As in, if they can delay a user an average of X minutes per viewing session, they save? Or perhaps more realistically, have they determined that the terrible recommendations do work for some metric?

It's not like they can be blind to the problem, right? Or perhaps they are lacking content for people like us?

While this makes a good narrative (and as someone who is fed up with laying down to get comfortable just to have to get up 3 minutes later to confirm I'm still watching Futurama again it's easy to imagine), I suspect it's less sinister than this. If anything I'd suspect it's more likely that they're trying to boost, or give a final effort, towards certain things. I'm sure viewing numbers play into their contract negotiations so if something isn't doing well it's an indication they need to ditch it or an indication they over-paid. So I wonder if instead of it being them trying to delay you watching something it's just them really hoping you'll eventually decide to watch it.

Yes, they should - this is a well-understood use case in recommender systems.

They'd run out of movies pretty damned fast if they did that...

These posts depress me. They are always showing off some cool algorithm that is oh so clever, and yet Netflix recommendations are just the same films over and over again. Amazon instant somehow nailed it - I keep finding new shows/films that are actually well aligned with my viewing history and rating. Maybe their offerings are more tailored to my tastes?

As a tangent, I've been meaning to start a discussion on how utterly unusable I find the Netflix interface. Maybe it is because I only just signed up and haven't been exposed to its iterations, but I am pretty sure I spent 10 minutes trying to figure out how to watch a particular episode. Likewise the personalised experience page sometimes disappears. Again, not wanting to compare too much, but for a late start to the game, Amazon I found pretty intuitive.</rant>

I have a theory that it is on purpose, they want the home screen to be filled with well-known movies and if you only have a handful of them at any given time it shows the same ones all over the place. Instead of seeing random not so popular movies and thinking of stop using the service because of few "good and popular" movies.

The core issue for me is that of the dozen or so shows and movies I would be interested in watching, on average 0.5 are available via netflix. So I spend a lot of time searching in futility for that show or movie I half remember wanting to see. During that search netflix is trying to recommend a bunch of other things to me and I find it annoying and distracting.

Also, could somebody somehow post a schedule of show/movie availability? Shows get added or removed on a regular basis and I always find out by accident.

(e.g. "Oxryly1! We do NOT have 'Guardians of the Galaxy' at the moment. We expect to add it in 4 months time. Thank you.")

</rant> indeed.

>Also, could somebody somehow post a schedule of show/movie availability?

Some websites have this information (such as http://whatsonnetflixnow.blogspot.com/)

Thank you for the that link... that's pretty much what I'm looking for.

You do realize this post was not about recommendations, right?

After using Netflix for quite some time, I have no evidence to support any hypothesis other than that their overall selection is secretly horrible and that they wish to conceal this however possible.

Evidence includes but is not limited to: Their user interface appears to intentionally border on unusable, a rambling Quora post by a C-level exec about how advanced search (a la Gmail) would ruin the world, and they've shut down their API, making it harder for third-party sites to index their selection, and allow positive UX to be possible thereby.

Yet they can't seem to figure out that where I am in several movies is after the credits have started, and no, I don't want my homepage to recommend I "keep watching".

If only Netflix could figure out the complex algorithm needed to prevent my children from being able to change their profiles from "teen" to "adult". http://www.huffingtonpost.com/knowmore-tv/how-orange-is-the-...

change the password next time they do it.

As far as we know, our kids have never done it. I just want to prevent the ability to do it. Working with netflix tech support multiple times, unless you live in Germany, you have to have all netflix profiles set to "teen". Otherwise your children can see shows full of sex scenes if they accidentally click on your profile button.

Surprisingly disappointed by this feature. "...Netflix records how much was watched and where the viewer left off." Isn't that kind of obvious?

I also wish there was a reset button for this. Sometimes I skim through a few seasons / episodes and when I go back and rewatch it there will be 3-4 episodes per season already watched and it will ask me if I want to resume, when in fact I don't remember.

>I also wish there was a reset button for this

There is. Click on Recently Watched, find the show of interest, click on the X to remove, and tell it you want to remove the whole series.

the 2 'low hanging fruit' solutions they can implement to drastically improve the recommendations:

1. use the 'percentage of series viewed' as a weight for the recommendations, so i don't end up with a list of shows similar to the one i watched only for 1 episode

2. filter out the shows/movies i've watched already

i can't even imagine why these haven't already been implemented, seem like such a no brainer.

try this, Netflix: if I've rated it, don't ever recommend it to be outside of "watch it again"

So much that. I watch a lot of movies. Now, most movies Netflix show me are movies I've watched. What's the point in that? Its infuriating.

So there's 2 billion hours watched per month, but a few billion "view" events per day? So let's round and say 100B view events a month, so 50 events per hour viewed? I wonder if this is because people preview a ton of stuff they don't watch, or are they recording pauses, or ... ?

good point, not sure. maybe it is from people hitting play once and binge watching a series consecutively "your program will start in 3...2...1..."

But since most episodes are at least a dozen minutes long, that should only generate like 10 events an hour eh?

Close, it's 60 events per hour. Ie. Once every minute an event is sent in saying "they're here now!". It's also sending a bunch of data about your quality of service and so on. This serves two purposes -- one is that in case you suddenly disconnect, at least it knows within one minute of where you left off, and two, it allows for monitoring the overall health of Netflix by taking the average quality of service and making sure it isn't dropping rapidly.

Ah, well that makes sense. Though could you not store last position on the client then just sync on connect? Like, if handling so many events was too much a challenge. Or are crashes that big of a deal? At any rate, I guess that allows you some leeway with the writes - a few minutes of unavailability doesn't hugely impact anyone eh? Fun writeup.

Many devices don't offer local storage to Netflix. Also, if your device crashes or your internet drops, there is a good chance that you're going to try and watch on a different device.

It makes for a much better customer experience if we don't have to rely on your device saving the information and calling home appropriately.

That eyeball in the circle of dots would make a good logo for an NSA project. They should borrow it.

That's the logo for Apache Cassandra https://cassandra.apache.org/ which is a highly available, fault-tolerant database designed for write intensive workloads.

The NSA probably uses it.

Is that a joke? "We focus on the user use cases" "We use Cassandra" "here is a simple diagram, which our intern has drawn in the last 20 minutes" That's not a tech blog post!

This is how a tech blog post looks like: https://www.dropbox.com/developers/blog/48/how-the-datastore...

We don't need to know that you record where the user left off, we want to know how you do it and why. That you store that information every non technical person can see after using Netflix the second time.

Your example blog post has missing images (404).

That's really interesting. The pictures still worked when I posted the link.

And yet they still can't figure out that if I skip the last 20 seconds of every episode in a 9-season series, that I've actually watched the whole episode.

I know it's their data. I know I'm playing in their sandbox. I know how much data a couple of terabyte hard drives can really hold.

But these kinds of revelations are still a solid 10 on the creepy scale nonetheless.

A 10 is overstating it, I think. Netflix gives reasons for why they collect the data (recommendations, continue watching cross-device, ...). Given the Netflix prize, I don't think we can be too surprised that they're collecting data for recommendations (or figuring out whether licensing deals are worth it). I'd give it a weak 2 on the creepy scale.

I agree. All the little value-added features they support that makes them different from competitors requires tracking more information. Being more useful with the same inventory is a matter of helping people utilize that inventory more effectively. That means knowing more about your customer. This isn't new with the internet, businesses have always tried to know and understand their customers. The internet just makes it easier to know more about individual customers, instead of broad categories.

Reply on news.ycombinator.com