I don't think their recommendation engine is working any less well; they've just made a higher-level strategic change in what it's trying to recommend.

Netflix used to want to give people the "perfect movie for them", expecting that eventually they would run out of perfect movies, and move onto just "pretty well-fit movies", and then run out of those too, and probably unsubscribe at that point.

Nowadays, Netflix seem to want to keep people subscribing by making them feel like they have a chance of finding the "perfect movie for them", but only after Netflix has recommended a bunch of those "pretty well-fit" and even some "not-so-well-fit" movies. This is "better" (from Netflix's perspective) for two reasons:

1. it gives people the impression that Netflix's library still has plenty of "perfect" movies to offer them (by stretching them out);

2. it gives variable-scheduled rewards, rather than constant rewards, making people more addicted.

I'm betting the new algorithm works much better at keeping people subscribed, despite seeming to work worse. Netflix is a very data-driven company, and wouldn't have switched to an algorithm that could be observed to hurt.

I think that can backfire. For a while, my "Top Pick For You" was the Emoji Movie. That just makes me assume that I've ran out of good (or even mediocre) things to watch on Netflix.

Add in the fact that their definition of perfect is evolving from "content people will like" to "content made by Netflix".

i think a simpler explanation is that they believe binge-watching long shows leads to better retention (they’ve said this) and as a result changed their algorithm to prioritize items which people keep watching, not items which people will rate highly.

Netflix basically wants to be a black box where people enter and are entertained for two hours a day forever. mind-expanding or emotionally affecting movies don’t really do that.

I have to second this. I used to watch some really "mindfucked" movies on Netflix. I think I watched The Cube, The Zero Theorem, and Brazil, partly because Netflix had recommended them to me. Now? Those movies don't even show up in the entire recommendation list. In fact, most of what shows up is series I know I won't like. I still subscribe but it's a letdown how the recommendation system incentives have changed.

I believe this was changed because of abuse.

Alt right people were giving Amy Schumer's standup 1 star. I remember watching it thinking it was good, not amazing, but certainly not 1 start standup compared to most of the filth out there.

Not just “alt-right” people. Quite a lot of groups have expressed their dislike lately. Personally, I find her incredibly bland, and I can’t say for sure that I didn’t leave one star for The Leather Special.

Could you provide a source that it was Alt right people doing this?

Don't be disingenuous, check your local alt-right source and see how they feel about her if you really want to know.

And yet Rotten Tomatoes had made it work for years.

1-5 star rating system works out when you have a lot of data; and don't work out at all when you have poor ones.

Now, that rating only makes sense given the clustered group of people with similar taste. It's also adds complexity of the clustering said group.

Furthermore, I think people generally are not always the best critic. I totally can picture people being precise in their rating getting the most of the the recommendation engine, and the sloppy getting poor recommendations.

In that case, I can't blame Netflix for wanting to trim down the complexity for making it easier for the audience and themselves.

I don't care about the sloppy ones. All I care about is how well it was working for me.

I personally liked star ratings more, but can see why they'd make the switch. Average star rating is basically a 1 dimensional model, but now what they're doing is communicating a prediction: how much will so-and-so enjoy this show.

A show can still have a lower match for all people (similar to having fewer stars). But I think I still like the (assumed) transparency of stars better.

Do they need a manual rating now with so many subscribers? If lots of people give up watching something after 15 minutes then it could be considered "not very good", and if people watch something multiple times then it could be "very good". Yes there are holes (people may give up on something good because they don't like the genre) but in general?

All Netflix movies are rated at 5.1, didn't you know? If you don't get the joke, you need to think more about UX.