Ask HN: How can Netflix and Amazon Prime be this bad at recommending content?
I’ve literally spent 20 minutes for the xth time in a row looking for anything that remotely captures me on both platforms and in the end did not watch anything. I’ve spent hundreds of hours binging series and movies on both platforms, but neither seems able to use this to suggest anything I like. I’ll probably cancel at least Netflix.
It would be so easy: - Store series I‘ve watched, show me new seasons when they release. - neither does it reliably. - Don’t keep suggesting things I have watched, except in a „watch again“ category. Especially using new pictures, getting me to click on the same stuff only to realize that I have seen this already.
Given the billions invested in these UIs, how can they fail so miserably? What am I missing? This makes no economic sense … or am I just a weird edge case, and this kind of UI works for the majority? If you spent 20 mins looking, it's not their recommendation system that's failing. It's that you don't like their content (or how their content is portrayed). The recommendation system is the thing that shows you stuff without you needing to browse for 20 mins. That's not true. Their recommendation system is failing. Often they will have content that's of interest to me, but instead I'm presented with all the popular titles that I've already watched and marked with a rating. I get the sense that they're A/B testing the wrong metric: time spent in the app. I'll often spend more time looking than a couple episodes of a show. I haven't been watching anything for a couple of years because I have two children and that limits tv a lot. However, the recommendation system still sucks. They haven't recommended me lockwood and co on Netflix, we stumbled on it because it was popular (and we binged watched it one night when both were asleep). So I do think it sucks. Recommendation seems to consider similarity, but not "quality" : they give equal treatment to their products, and the consequence seems to be that they show often garbage mixed in with good stuff... It's exactly the issue: it recommends based on "same style of what you watched", not on "this is high quality". I watch high quality shows only - anyway I don't watch much. And I have the exact same impression, it doesn't suggest me quality shows. Yes that's the same thing I noticed. Interestingly, Steam seems to do a much better job at recommending OR my ability to filter out good/bad videogames is much superior. The problem with a movie is that you can't see a part of the plot for some of them or it would spoil it, while for videogames you definitely want to look at the gameplay (at least for me) before buying. That's simply not true. There is plenty of content I like, but I am not getting it recommended. Often I find stuff on Reddit or other forums or through my own research then look for it on Amazon or Netflix and often it is there, but was never recommended to me. Also paradox of choice / overchoice phenomena, where the more options you have, the less you feel satisfied with any of them. I don’t think its applicable here. I personally have many platforms and options where I can find most movies that I want to watch. With this many options I have no trouble finding movie I’ll almost definitely enjoy watching. This process for finding right movie takes effort and many aspects, while the streaming service only has my watching/browsing history and rating system which I don’t bother participating in. I think it might get better if they become more intrusive, though. Yeah, the problem isn't new at all. It was much easier to choose a film to watch in the cinema (about 5 options) than to choose a film from the video store (500 options). Considering you wouldn't feel satisfied most with only one option, what will be the optimal number of options? > Considering you wouldn't feel satisfied most with only one option Not true, the opposite often happens when you don't really have a choice and are mentally exhausted. If you're expecting choice and don't get it, then of course you're not going to be satisfied. But the lack of any choice (other than just leaving) can make certain people in certain states more satisfied than anything else. Legacy broadcast radio worked on this principle for a long time, and still does for people who just keep their car radio on a single station and let the DJ or talk radio producer make all the decisions. > what will be the optimal number of options? That varies significantly, both between individuals in terms of the average tolerance, as well as within individuals in terms of their present psychological state. Plus if they‘re constantly on there they may have watched all of the content they like on these platforms. I have considered this, and there’s some truth truth to this, that I have watched all from my favorite categories … but then I do discover whole seasons of things I haven’t watched but liked previous seasons of, and whole series when digging around long enough … it just feels increasingly difficult to efficiently dig. Maybe recommendations are harder than we think because our tastes are too arbitrary and unpredictable. Maybe overlapping with someone else's tastes actually predicts nothing about whether there are any more places we will overlap, so it's not the simple data processing problem that we wish it to be. I've tried all sorts of "people who liked X also enjoyed Y" book and movie lists and it never seems any better than a generic list of decent books and movies. One of my favorite movies is The Arrival. To me, it's one of the best love stories ever told. Yet when I meet other people who really liked the movie, that wasn't the main draw for them. Perhaps it was the intrigue of the time travel or watching interesting characters navigate a complex conflict or perhaps they liked the provocative questions that it explores. Perhaps two people liking The Arrival says little about what they have in common at all, thus there's actually no data to go off beyond recommending both parties more critically acclaimed movies. I suspect that this might be the cold reality of recommendations. Belgium Canvas has several superb movies and series, every week, year-in year-out. Humans select them, not software. Selecting movies/shows is a profession and I’m positive that many people can learn it. When I meet people with seemingly the same taste, usually we exchange our trophies, eg my latest Your shows all seem to have a 7.5+ IMDB rating which means pretty much everyone should probably give them a chance if they like the genre/synopsis. (I certainly will, thanks.) The idea of a recommendation engine in these comments is generally a holy grail that can predict shows you will like personally that aren't necessarily well-rated by other people. Shows that don't simply show up in the platform's "critically acclaimed" section. Though your post is a good example of why word of mouth recommendations are important. Netflix could have the best recommendation system in the world yet it cannot recommend any of the great shows it doesn't have to show you. Also, what appeals to us is heavily dependent on our mood and situation at the moment. There are many times when something appeals to me heavily one day, and not at all another day. Even on those different days, I can still recognize a show that I might like to watch on another day though. On youtube I put those in my Watch Later list. I have seen it getting worse and worse over the last years and also I have watched so much on these platforms that I‘d think the signal should stand out no matter the underlying reason for it. As someone that has worked within one of these orgs, it's because the recommendation system has been captured by contract negotiations and forced merchandising. You are not seeing things they think you would like. You are seeing the things they want you to watch. I find this very interesting that the orgs allow this to go as far as making my experience so miserable that I consider leaving the platforms. I guess it’sa slippery slope and provably also metrics that are chosen to hide my increasing disinterest … everything old is new again https://en.wikipedia.org/wiki/Payola Totally agree, that's why I gave up on it myself and created an aggregator. These are other similar filter-based aggregators (mostly Netflix) https://pastebin.com/raw/ZpsCywHe if anyone is interested. This looks good but the anime genre doesn't bring up much. That said, not a lot of reviews of anime are on rotten tomato either. For example, The Great Pretender is at 100% from critics but doesn't show up at all here. This looks pretty good, thank you for sharing. I have a feature request: ordering by number of seasons. Very cool, just discovered something to binge :) thanks! What is min corn and min strawberry? Clicking one of the results, it seems to mean Average Tomatometer and Average audience score … both from Rotten Tomatoes. yeah i had to change those unfortunately because i got a dmca for using the signature tomato and popcorn icons. Looks like corn is the audience score and strawberry is the critic score, sourced from rotten tomatoes. And yellow numbers seem to be Certified Fresh. IMO Netflix and Amazon Prime's problem is they don't surface their good content enough. They mix it with their shitty content. They should make it stand out. HBO are good at this. When they make something good, that they know is good, they will rub your face in it with large high quality images. I don't have time to shift out the good from the bad with Netflix. HBO keeps tying to get me to watch Velma. It's hot woke garbage wrapped in bad story telling. Something else conspicuously missing from these services is a checkbox to make it never show me something again. Why do I have to scroll through an endless line of "nope", "nope", "nope", "definitely not", "no" every night, then look at all the same stuff tomorrow? Does someone get paid when I have to sacrifice a brain cell for each of these minor decisions? Netflix allows you to down vote the content with "Not for me" (thumb down) option. Then it stops suggesting it. The sneaky part is that the option itself is hidden a bit, under "thumb up" button (which expands to three different options if you hover your mouse over it), so I guess we've found another dark pattern :) The real problem is that each platform just doesn't have that much content. Everyone wants their own platform, and now the content is so fragmented that if you noped out of titles for very long you'd realize they have very little else. But I have the same experience on youtube. > Does someone get paid when I have to sacrifice a brain cell for each of these minor decisions? In many cases, probably, either directly or indirectly. Payola is huge in the music industry and it’s why, despite never having heard of Harry Styles before, Apple Music started pushing him into every corner of their app for a few weeks after his last album release. I can understand surfacing artists who are likely to be popular with the general population. But this was clearly more than that. I’m sure it’s the same with video services. They’re paid to promote specific shows. It probably also costs them less to show you their own shows, so those are probably surfaced more frequently than shows from other companies, etc. I do believe they have some kind of "monetization" going on regarding what they recommend you. Even YouTube apparently. They suggest the same thing over and over under "different" categories. I they go try to force "popular" content on you. One of the reasons I made this joke/rant project Because priority number one is getting you to watch their originals, so that you will keep subscribing as the market for movies produced outside the streaming services gets more competitive. The big events gets featured on the home page, for anything you are probably better of searching for directors or reading reviews, and then finding where you can see it. Yeah I’ve noticed this as well, and I have liked quite some of their original content … But if it gets to the point that they’re failing to keep me as a custoner at all, something must be amiss … A big problem on Netflix is that they don't use the original poster or description for movies, instead coming up with their own. I find that these generally don't do movies justice. Not only that, they dynamically change the poster to try to make it more appealing to each person and it still fails miserably. They wrote a blog post (that I can't seem to find) a few years ago about how they A/B test their thumbnails. I remember reading that they have an algorithm that picks the "best" (colorful, cinematic, attention grabbing...) picture to show in the UI from the content itself. That is why sometimes the poster used represents a very minor character instead of the main lead. Similarly, I'm not sure if I like that they play a short scene of a movie when you hover vs. showing the original trailer. Sometimes I like to watch familiar favorites but they just show me one short scene and it doesn't really draw me in as I had hoped. You can turn that off, but you have to go to Netflix settings on a web browser and turn it off for each profile. Because reasons. Maybe the problem is that they don't carry more of the content that you like? Their goal is to keep you subscribed, not to keep you entertained. They know they don’t have enough premium content to keep you as a subscriber if that’s all you watch. That kind of content usually costs a ton to create. So they need to have you watch filler content as well. And just like social media companies have taken advantage of the slot machine effect to increase addictiveness, so too do streamers. If everything you watch is good, you’ll end up watching less and be more likely to cancel. You’re assuming that the goal of a recommendation algorithm is to recommend the thing you’d most want to watch at that specific moment. It’s not. The goal is to maximize their retention. It’s amazing how many of life’s annoyances become instantly explainable when you realize that your interests are not aligned with those of the companies you’re dealing with. Your happiness will always be a secondary concern to their profits. Human beings cannot make good recommendations most of the time. What makes you think an algorithm can? My major problem with almost all recommendation engines and add-ons is that most of the time I want a recommendation that is orthogonal to the last thing I watched. If I just watched a Japanese yakuza crime thriller I don't want you to recommend another Japanese yakuza crime thriller. Give me something as good as what I just watched but different. But that's me. Many people just want to watch Hallmark Christmas romances until their heads explode. That's probably why I still use Criticker. Because it gives you one choice across multiple genres. My feeling is that Payola is most likely the reason why the recommendation algorithms are so bad rn. Yeah. I especially find it interesting when Netflix says something you couldn't pay me to watch is a 92% match with my profile (or however they phrase it). But part of the problem the algorithm has is that content is of very uneven quality. I look at very recent things that, a few years ago, I would have wanted to see, and scroll on by. The first Avengers movie? Kind of rushed but acceptable. Thor Love and Thunder? Garbage. (Not Netflix examples, just examples) We want to see new content but have learned from recent experience that it's not worth the trouble. The algorithm simply can't understand that. I got a tip from someone here and shifted to tracking what I want to watch with a seperate system trakt.tv. This lets it recommend across different services and removes incentives to promote their own content (I'm assuming someone at Amazon and Netflix is trying to goose their own stats by boosting the content they self create, or that they pay the least for) A link (and a free recommendation for an interesting show): https://trakt.tv/shows/giri-haji It lets you do nerdier searches like by director, writer, actor and similar to openlibrary.org it has some weird old stuff too. FWIW, I typed in track.tv into my browser and it loads up with a blank page. I have JS turned on, so that’s not the issue, but I do also have ad blockers. This is with Orion on the iPad, if that makes a difference. Sounds interesting, but if I can’t physically get to it, it’s going to be hard to use. It's trakt, I typoed the name when writing it but the link I added above should work. Ah, thanks! That did the trick! All streaming platforms i tried have this problem. Maybe the ranking of shows is commercially funded? Also annoying: netflix shows multiple cover images. Most annoying automatic playback of preview (including sound) in the menu. It doesn't need to be explicitly funded, different licencing costs can be enough. It could be cheaper for Netflix to show their own productions than pay for a stream of licensed content. You can disable autoplay previews for all devices on Netflix' website in your profile. This is the real answer. Why do Apple and Amazon have search “ads”? They’re specifically trying to push content your way based on the how much the content creator/producer pays extra. Even grocery stores do this - aisle placement is often something the food distributor pays for. Read an article about how Netflix stages in house CDN "boxes" directly in key ISP locations, directly on the switch. Certain high ranking shows are spread through the CDN for responsiveness. Some of the recommended shows are the ones staged close to the watcher to eliminate lag. https://about.netflix.com/en/news/how-netflix-works-with-isp... The article you linked to doesn't say this at all. It says almost the exact opposite; they cache shows based on the algorithm, not that they recommend shows based on what's cached. What is the relevance to the OP’s question? I think Netflix needs an "I really don't like this or programmes like it" option. ...come to think of it so does YouTube. I'd love to be able to search for a particular genre and exclude everything that comes up. Hitting "not interested" seems to work for me. Not for me sadly. Like a bad case of the pox no matter how many times I click a "thumb down" or "not interested" (including the explicit "I. DON'T. LIKE. THIS." option in "Tell Us Why") certain things[1] will be back in my recommendations again in a couple of days. 1. It doesn't matter who likes what. Everyone's had enough of something. In my Netflix I can thumb down content. As others have mentioned, part of the goal is to induce you towards the service's self-made programming. They're rolling the dice that you will get hooked on something if they show you enough options, and your time spent searching for content (as long as it doesn't lead to an unsubscribe) is cheaper for them than time spent consuming content. FWIW, Spotify has similar issues in that they can't seem to figure out that new music from artists I'm "following" should be a slam dunk for the "new music for you" list. Time searching is cheaper then time watching is an interesting thought … I do hope this is not a metric where my searching time is seen as positive … because that would be a very short sighted metric Is it thinkable that after "hundreds of hours binging series and movies" you have simply exhausted the options? Me, I haven't found a second The Wire yet, no matter how often a recommender claimed I have. Have you seen https://www.hbo.com/we-own-this-city Baltimore and the same Wire producers and actors. The only con is it is only one season. Try True Detective For me personally, the recommendation engines choose the wrong aspects of shows that I like. They focus too much on genre, and not enough on writing qualities, variety, surprise, etc. Just because I liked one “gritty suspenseful murder mystery” (or whatever the genre is) does not mean I will like another! The execution could be wildly uneven in different programs that fall in the same subject matter bucket. I’d rather see a recommendation that can figure out my tastes across genres. You are asking for recommendations and they probably don’t have anything. They’re not going to say that. So you get things you don’t want and things you have already seen. If you are looking for movie recommendations, nothing beats movielense https://movielens.org I have thought for years Netflix (for me, at least) would benefit massively from a "never show me this show again" button. I continually see the same bunch of shows when browsing that I will never, ever watch. They take up space and add to cognitive load and ultimately just annoy me each time I see them in the list. I'm sure this weighting would also be very useful from an algorithmic recommendations view as well. For Netflix's DVD service, pre-streaming, their recommendation system was awesome. You know, the one where they had the contests and paid $1M to any group that could do better by some percentage? IMO, they don't use anything like that now for streaming. It's as others have said: this is what we want you to watch, not what you necessarily will like. For Amazon the issue for me has always been that aside from the things I shop for myself I get asked to buy things for other people or presents for family members Amazon doesn't know who I bought something for so making sense of why I have bought 4 padlocks (my father in law), a set of sauce pans (wife), a light novel (myself) and rechargeable batteries (myself) Yeah, but I consciously watch things my wife likes with her on her account exactly due to this ;) Mass media platforms are very shallow when it come to "managing personal information". Netflix is a disaster in that regard. and yes this kind of UI works for the majority: To be honest, the average person, normally content with switching on a TV, couldn’t care less. Recommendation algorithms maximize the time you stay and interact with the platform.
As long as you stay on the platform and consume content its working. Its a bit the same as dating algorithms. If the algorithm was to good there would be no paying users of the service. When I was growing up every kid watched more or less the same cartoons and played the same games. Life was simpler. Now as an adult there's too much out there and too little time. I suspect that this is a big part of why people feel there's nothing good to watch. But if there was too much, shouldn‘t I have too many good option rather than too few? The good is mixed with the bad and it takes more work now to find the good. So perhaps the OP is correct in that the recommendation systems could use some improvement. I actually went back to pirating movies/shows and host them on Plex. I sure have a much smaller collection than any streaming service, but it's actually all content I like, and I actually watch stuff instead of just browsing for 20min. Maybe they push content that is cheaper to license optimizing for some sort of license-cost-per-viewer-hour metric contrasting whit YouTube or tiktok is harder to Know why people like series ,less content, series become good after 40 minutes, people let their content run, do you want a background series or are u seeing in the momentfully ckncentrated, lees feedback look all the way down Edit: are you seeing alone or whit other Honestly, I have yet to see a recommendation system that works very well. PopcornTime Stop relying on recommendations and figure out some humans that curate its also possible they wont show you content you really want because its licensing is expensive, and instead they nudge you to content with higher % rev split Because it's profitable to recommend shit. maybe you don't want to watch tv These services get paid to promote content. Did you think they are going to promote legit content? Netflix gets paid to promote content? That doesn't make sense. They buy or license content, and their incentive is to get you to like what they have to offer in order for you to keep paying them a subscription fee. Why would they pay someone for content, only to have them then pay them to promote the content? Spotify does this too (paid recommendations). "Help me find things I like" is basically a mandatory requirement for anything whose catalog is too large to search exhaustively — otherwise there's no point to them. Doing this well seems to be surprisingly difficult for everyone. Facebook has shown me local news in Florida even though I'm British, live in Berlin, and have not even visited that state. Twitter thought I was interested in baseball, when I don't care for any spectator sports. YouTube is mostly OK for the long form content, but the shorts are 98% useless, and the adverts are… well, the current ad for me in the app is "Click this video = $1000" ad by "Beast Promo" with a cartoon that looks like MrBeast, which absolutely screams "scam", and last week it was something that looked like an anti-LGBT conspiracy theorist but my German isn't good enough to be totally sure. Netflix, Disney, and all the rest? IMO Netflix is the best of them, but still not really all that amazing — the bar it passes that the rest fail is excessive focus on their own content. Maybe but it’s in their best interest to keep me watching… surfing and leaving without watching should be a very strong sign that something is amiss
This is going to hurt (series) - 2022 - UK
En man som heter Ove - 2015 - Sweden
All That Breathes - 2022 - India
Jeanne Dielman, 23, quai du Commerce, 1080 Bruxelles - 1975 - Belgium