Hacker News Reader

Comment by isbjorn16 | Hacker News Reader

My problem with Netflix isn't the slate of shows and movies it has in its catalog. It's definitely not as nice as it used to be, but there are a jillion and a half shows and movies out there.

My problem is that Netflix has taken note that we watch anime a few times a week, which doubtless means we want every single band/carousel/whatever to be about anime. New Anime, Critically Acclaimed Anime, Popular Anime, Watch it Again Anime, Anime Where Characters Wear Hats, Anime Where Characters Float in Hats, and Anime About Hat Making.

For the love of all let me get some other options in the fucking list before I lose my mind. I like comedies, action movies, sci-fi, documentaries, and historical dramatizations or historical fictions too, you know. Not everything has to have a talking cat in it.

View on HN

Once upon a time, Netflix gave away $1 million to contest winners in an attempt to get the best recommendation algorithm they could ( https://en.wikipedia.org/wiki/Netflix_Prize ).

Now the recommendations are just not that great. It's like they're not trying too hard because they are comfy in their market position.

I have never watched any anime at all (really not my thing), but Netflix keeps recommending it to me despite the fact that I haven't taken the bait once.

Also it recommends shows to me that it should know through history I've already watched every episode of. Maybe that's on purpose since people watch things again, but I still don't find it useful.

Also it leaves stuff in "Continue Watching" for like a month after I give it a thumbs down rating and remove it from "My List". That should be a strong enough signal to act on.

And it seems to recommend stuff based on stuff that I only "watched" because it auto-played. That should not be a strong enough signal to act on.

The overall point is they seem to have several obvious opportunities to be smarter with content discovery, but they're not taking them.

Wait where's the thumbs down rating?

im pretty sure they removed it

edit: i could be mistaken though

Next to the thumbs up rating (they only have two possible ratings). It might be confusing as you sometimes (at least on iOS) have to click a thumbs up icon to be shown the actual options with the thumbs down?

I suspect they don't have it at all on their Roku channel :( I'll look again tonight, but if I can start downvoting these things, I can at least feel like I'm screaming "no" into the void, which is better than screaming "no" internally. Nobody wants to live through an "I have no mouth and I must scream" nightmare, after all.

Netflix has thumbs down on Roku. That's where I use Netflix primarily, so I'm sure I've used it there.

Perhaps their recommendation engine is working as designed, but it's tuned for some other metric besides maximum viewer enjoyment. For instance, rankings might be influenced by the royalties they pay, or the likelihood that you'll recommend a Netflix-only show to a non-Netflix-subscribing friend.

Their discovery algorithm is hideous and on most of their TV streaming apps the search is shit.

They didn't end up implementing the winning algorithm due to how complex it would be to run at scale.

Content satisfaction was more important when they ran a service where you spent several days waiting for and returning a disc.

The metrics shifted. Subscriber retention just isn't that tightly correlated with them identifying hidden gems for you, that may be an externality we've just lost.

There used to be a lot of competition in this space, for a few years you could go on free sites designed only to give you targeted recommendations. I think Jinni was one.

They were just training their engines and became ad networks.

One other factor, not really anyone's fault -- the massive increase in high quality scripted content has made search less critical. If you fail to find hidden gems in 2003, you get Pirates of the Caribbean sequels and miss out on Oldboy or Goodbye Lenin. Today if you pull up any random streaming service you'll probably find something that's good enough, even if not great.

It's a little disappointing, maybe there's room for a p2p ML recommendation engine in this space for people who really want to optimize what they're consuming. I'd be into it.

I often enjoy Netflix recommendations that look like shows I wouldn't normally try.

Or maybe the Netflix Prize was just a PR/advertising exercise. $1M is cheap, all things considered.

I have this problem with ML and recommendation systems in general. I hope this fad goes away soon and people go back to making simple, understandable, show-me-the-newest-stuff feeds.

Right - I just feel like it takes things entirely too far. Not every signal deserves a boost that takes a certain genre of suggestions up to 11.

These recommendation systems need to be a bit less zealous; I like black pepper, too, but that doesn't mean I want a huge bowl of black peppercorns in cream sauce for dinner.

They may be doing the math on extremes in categories where relative quality is good given their current catalogue.

People tend to look up movies not by theme, but by actor or director. Could Netflix be okay with categories that don’t fill up entire screens?

That blows my mind; I never would have considered people would focus on the actors or the director. Until you said it I don't once thing I ever even considered the possibility! I'm kinda floored right now.

Actors or directors are a decent indicator of quality.

Recently I’ve found that old actors in mid-budget productions have the best overall quality.

Netflix didn't always have this problem. They seem to have changed their cold start model within the last 2 years.

I think the issue lies with there being a single recommendation tab that gets filled in without enough variety.

One reason I appreciate broadcast TV is the curation aspect of their limited playlists. The streaming recommendation systems are very lonely - I know they are algorithmic and personalized, and that nobody else will see the same outcome. With broadcast, a human does that work and it is shared among viewers; e.g. some show always airs on Thursdays at 8. Then there is the seasonal cadence, i.e. summer shows, fall/spring shows.

With algorithmic systems it sometimes feels like reading books written by a Markov chain toy.

Better than Hulu that now shows us every commercial in Spanish. No one in my household speaks Spanish. The only thing I can guess is that I know ONE TIME my daughter accidentally selected the Spanish translation of one of her cartoons.

This is a guess but perhaps your subtitles are in Spanish, and you never use them so they are hidden?

Or perhaps some global preferences setting hidden in the shadows to the left somewhere

The New York Times app shows me ads in Spanish. Don't speak Spanish. No one in my house does.

Well, time to pick up duolingo's Spanish course...

When life gives you lemons…

You should petition MaxMind to fix their mapping of your IP address to city/country.

Year ago, Netflix has a "Not Interested" option.

If Netflix was recommending something that you didn't care about, you could use "Not Interested" to make it not show up in your list anymore.

Now, it just keeps recommending these same content over and over again. There's no way to say "I don't want to watch that". There's also no variety to the suggestions.

I agree that "decision fatigue" is a big deal for me.

Recommendations used to seem to be affected by your ratings. That may still be the case to some extent, but Netflix's preferences seem to be way, way more important now.

That aside, I don't live in the USA so it would be nice if Netflix stopped claiming that USA content was local, and non-USA foreign. It's not just their recommendation system that is poor, their general categorisation also is.

It seems like there is some pitfall in making a recommendation algorithm that more and more people are falling in to.

I don't watch TV or have Netflix, but I listen to streaming music, and I noticed something weird. I used to listen to Pandora, and it recommended a lot of different artists that I liked on hearing them for the first time. It single-handedly gave me a taste for 21st century music. But there wasn't enough music in their catalog, so I switched to Spotify. Now Spotify will play more of the songs by people I already like, but very nearly everything it plays that I didn't find on Pandora I hate and it keep playing the same things over and over no matter how quickly and frequently I skip past them.

It's hard for me to believe that they could be this bad and unresponsive unwittingly, but I wonder if perhaps there is some kind of naive algorithm that is failing.

...you may say, why don't I explicitly mark songs "like" or "dislike", but I mostly listen in my car where I can't do that.

One alternative hypothesis is that different people use Pandora vs. Spotify and the ones on Pandora have better taste, but I can't believe that's all of it.

Its less risky to give you things you already like than to try and guess if something new will be appreciated. Its the same reason hollywood keeps doing remakes.

If you only go by statistics and don’t think through the “what if everyone did this” question then giving you the same old content on repeat will have better metrics.

Right, it feels like things are optimising purely for “next click”, i.e, what will keep you mindlessly in flow, forgetting the time and consuming more content.

So if I’m watching a video about TIG welding say, to optimise “next click” I suppose I should be recommended mostly more welding videos, with a smattering of engineering and finally a bit of clickbait mixed in case I’m starting to get distracted.

If services are optimising for “keep going right now”, they might be modelling as something like a markov chain of “clicked on next” and that’s how we get stuck in this kind of short-sighted local optimum.

Youtube does this too, and its really killing the recommendations for me. Same thing with Quora, I used to browse Quaora all the time, until the content ended up just being the same. I think in the aggregate these algorithms are driving more engagement, and pissing off some people, but not enough for them to change the product.

Pandora has almost no variety at all. Even a playlist like "90s Alternative" which should have thousands of potential top songs plays the same songs every time I listen

Google Play Music has been the best for me in terms of variety, but it's buggy as hell. It freezes up if it can't play an ad for some reason, won't automatically play when bluetooth is connected (like Pandora does), and other annoying things. (I ride a motorcycle so I can't mess with my phone if something goes wrong)

Pandora has almost no variety at all. Even a playlist like "90s Alternative" which should have thousands of potential top songs plays the same songs every time I listen

If a streaming service didn't order automatically generated genre playlists by popularity (with a little randomness) people would stop listening to them and start saying the playlists are full of rubbish songs they've never heard of.

I don't think I started with a playlist like that. It's been a while, but I think I started with specific artists/music (albeit from the 90s) that I liked. I don't think you can expect individually satisfying suggestions if you request generic music in the first place.

Yes! That has driven me away from YouTube and Netflix recently. I know there is an ocean of content out there but when I sign in all I see is the same dozen shitty recommendations that I have no interest in! I watched one comedy special a year ago and my feed is half stand-up comedy now.

On PC you can click on a icon near the video on the home page and say that you're not interested.

This happened to my friend too. I shared a link with him of a Youtube guy who analyzes classic rock songs, and my friend's entire recommended feed was overtaken by this guy for weeks. My feed isn't nearly that sensitive. I think you and him are in the same experimentation bucket or something.

or your kid accidentally starts playing one of the "mock buster" movies. like the absolute garbage take off of Pacific Rim called "Atlantic Rim."

Now my feed is full of those. I wish Netflix would stop buying those movies.

To be fair to them, everyone is failing at this. Recommendation algorithms, once held up as a crowning achievement of machine learning, have turned out to be terrible at their jobs. Amazon, Youtube, Netflix. None of them have it figured out.

YouTube seems to be a lot better than the rest. I consistently get good music videos queued up, it is great for discovering new music

YouTube seems to give me clickbait crap with high view counts. Yes it may somewhat be related to my interests but not really.

Spotifys algorithm is amazing, I'm constantly finding new indie bands that I love on it. YouTube is also really good. Netflix is terrible for me, I don't think it's ever recommended a show I've loved.

Huh. My experience is the opposite. Somehow my Discover Weekly was mostly mariachi bands for a month. Finally told it I didn’t like enough of them that it got the hint. Now it’s overloaded with Scandinavian adult contemporary or something. It’s like all the other algorithms from one I’ve seen: listen to something accidentally that’s in a niche and it just loads up on recommendations from that niche.

For me it's comedy specials. I don't even like them that much, but they have been plastered all over my home screen for nearly a year. Since they took away ratings and reviews I have to use other sites to try to find good stuff to watch.

I've been watching a lot more Youtube recently oddly enough. Youtube does a better job of giving me a mix of videos for my interests, and there are a number of great vloggers out there across a lot of my interests posting weekly+ content.

It become more or less just a catalog, often times spent close to an hour just trying to find something might be interesting (not totally shitty), and that's pretty much it: couldn't find anything, just browsing their catalog, with time wasted.

This is the classic article on that problem:

My Tivo Thinks I'm Gay https://www.wsj.com/articles/SB1038261936872356908

That article was a wild ride back to 2002. From the phrase "techno-profiling" to Amazon recommending "Slave Girls From Beyond Infinity" to Bezos during a public demonstration to WSJ writing "Netflix.com (netflix.com)" so the url was clear to anyone whose browser didn't convert case before making a request. In some ways technology has made a lot of progress and in some ways it really hasn't changed a bit.

Reply on news.ycombinator.com