Is Netflix recommendation engine the best there is? | Hacker News Reader

Is Netflix recommendation engine the best there is?

1 pointxsenna8 comments

So, I just started using Netflix after a 2 years break (it did not have a good library available in my country). And I see that the recommendation engine is still a joke for anyone with sort of advanced taste of movies / tv series. Even the Spotify recommendations are light years better (yeah, it's not 100% the same for music and movies, but pretty similar). Here are some major problems that I've encountered:

1. It shows movies that I already saw as recommendations. I personally rarely rewatch movies, and definitely do not want them to be top recommendations. Pretty sure, that many people are the same. Why it does that, I don't know. Also, there is no way to indicate to the system that you watched the movie but did not particularly liked it or hated it.

2. One of the top movies that I should like, according to Netflix is Uncut Gems, which I really hated, because it has an extremely unpleasant protagonist and a boring, uneventful story. There were only three original things in the film: cinematography, music and random "bad luck" events that had profound impact on the protagonist. Netflix thought that I should like it because I like Dark Thrillers, which is true, but I like Dark Thrillers that come with a good story. So if I like Fargo, does not mean that I will like Uncut Gems.

3. This leads to probably the biggest issue, which is: not enough dimensions on which to make a judgement. I would love to watch all movies with a strong original plot, but it is not one of selection criteria. I love watching most movies of my favorite directors, so (say) Tarantino, Lynch, Cronenberg, Nolan, Boyle, Villeneuve, and many more would be a 100% match, and the algorithm should be able to perceive it. I love good foreign and sci-fi movies, but I do not see them in the list, most probably because my taste is a bit more esoteric than the tastes of the crowd.

The original Netflix recommendation challenge was floated back when Netflix was a "movie by mail" service where if you got a good or bad movie the situation would take days to play out. In that case the difference between 90% and 95% accurate is something the customer can feel, is good for the business, etc.

Back then Netflix was buying retail DVDs so they also had an unlimited catalog: they could start stocking a movie just by buying a few disks.

They negotiated with Hollywood for the rights to run a streaming service which was structured something like "Netflix pays Disney $0.80 a month per subscriber."

At this point the "Netflix optimization problem" is: "buy content that will keep our customers coming back for the minimum amount of money." Netflix withheld data about which movies customers watched from the studios because they didn't want studios to know what they thought the movies were worth.

Then Netflix started to make its own movies because, once the number of subscribers is large, it is cheaper to buy then to rent.

The new optimization problem is still "minimize costs" but now Netflix is involved with the production of movies from the very beginning and will produce movies based on what plays well (or they imagine plays well) with the Netflix audience.

Is it good?

I don't think so. I saw cable channels become fat but disconnected from the consumer because they had a business model where they got paid whether or not you watched. In the "GAME PASS" future you can play any Assassin's Creed game you want for $10 a month -- I'd rather decide which games I buy myself instead of have it be done by some insiders in a smoke-free room.

Agree, it seems that the business model change of going from DVD by mail to streaming are much more profound, than what is usually discussed. With everyone going into streaming, it looks like at least in short-term perspective we are moving to a cable network of streaming channels, with 4-5 networks being owned by a single household that wants to see all that's good out there.

Netflix is incompletely understood by most people because it's growth unfolded over a long time and it went through a few sharp pivots.

DVD-by-mail is gangster in terms of intellectual property, but the studios could do nothing about it (e.g. it is like a book)

What Netflix gets for it is a huge and loyal customer base.

The studios are then willing to license the streaming rights because they get more that way then they get in DVD-by-mail.

Netflix is busy making streaming delivery work.

As the scale changes Netflix goes from rent to buy, from David to Goliath. So far Reed Hastings has been thinking a step ahead of others in the industry at each juncture.

4. Another dimension is how many stars the movie got, or in Netflix's case - number of thumbs up and thumbs down. Not sure how many people vote that way, but ideally it will be something like an IMDB score. I am quite prepared to give a try to what most people consider really good, but IMBD score is 0-10, while thumbs up is much more ambiguous and it probably works against movie buffs, who will thumb up only the best, but will be drowned in tastes of the average person. A movie that was voted higher than 8 on IMDB by a lot of people is highly likely a good movie that should be watched, regardless of genre, language, etc. The only problem IMDB has right now is a mass of Indian and Turkish voters (not sure if it's bots or just a flashmob crowd) that got quite a lot of their movies in the IMDB top. However, Netflix does not have this dimension at all. Movies that were highly rated by critics or curated lists also don't show up and so the judgement goes towards a common denominator, instead of receiving a bump from critics.

5. Finally, an issue that becomes apparent after some time, is just a really tiny library of movie content. If the info on the Internet is correct, there are no more than 5-10k movies in the Netflix library, this is nothing compared to how many movies are out there. Whole genres are not represented, where are the hardboiled Hong Kong action, foreign Sci-Fi, classic oldies, etc. It is not a direct problem of Netflix recommendation engine, but a limitation still. Again, if I compare Spotify and Netflix coverage, Spotify fares much better, with no more than 10% of artists that I listed to no listed. Of course, Netflix tries to switch everyone to watching just more of their own content which they can distribute without issues or additional cost worldwide. But if the market becomes fragmented, the movie discovery function diminishes in value.

It never applies to me.

Isn't the problem that they game the recommendations too much? They obviously emphasize NF "Originals" and probably content they are getting on better deals. Regarding point 1, they very often plug watched titles again and again (I'd prefer them just being hidden), I think this is because of their rather crappy catalogue (probably also the reason they show you different posters for the same title). I've never gotten the impression they actually have 5888[0] titles.

[0] https://unogs.com/countrydetail (USA)

It used to be better years ago IMHO. Then they reduced your opinion possibilities to up/down and the recommendations seemed to get worse.

I also think it sucks at its current form, but I am surprised that this is supposed to be the best Silicon Valley can offer.

Yuck, No ! I hated it.

Reply on news.ycombinator.com