Netflix and the Napoleon Dynamite Problem (2008) • Hacker News Board

Re: Netflix and the Napoleon Dynamite Problem (2008)

#41

Post by iak8god » Wed, Jul 22, 2015, 6:53 PM UTC

mrspeaker wrote: post #14

"If you use a computerized system based on ratings, you will tend to get very relevant but safe answers"... I call this my "Radiohead metric": for a given music recommendation system, how long do I have to use it before it tries to recommend me Radiohead. One of the problems I find in music algorithms is they can't know how I've changed - suggesting music based on what I play or input - but that's stuff I liked in mi…

> I call this my "Radiohead metric"

I don't use Pandora very much any more but when I did I'd just create a station based on a single artist and then usually listen to it without further interaction. All my stations would converge on Radiohead within ten - twenty tracks regardless of starting point: Joy Division? You must like Radiohead. Passion Pit? Radiohead. Billy Holiday? Radiohead. It was amusing but also more than a little bit annoying, and part of the reason I moved on from Pandora.

Re: Netflix and the Napoleon Dynamite Problem (2008)

#42

Post by wtbob » Wed, Jul 22, 2015, 6:55 PM UTC

quotedmycode wrote: post #2

From what I understand, they never implemented any of the algorithms, because their existing one was good enough, and the premise has changed. It used to be, you'd make suggestions so people could get DVDs sent to them, and you'd know they would enjoy it. Now with streaming, the cost to send something you don't like is cheap. So the recommendation engine just has to come up with something you'd try to watch and perha…

> So the recommendation engine just has to come up with something you'd try to watch and perhaps enjoy.

Heck, if it did that I'd be happy. I don't even know why I still have my Netflix subscription. The selection is basically as good (bad) as Hulu and Crackle, the price is higher and the recommendations are laughable (right now they're pushing: Velvet, a TV series about a Spanish fashion house; Sense 8, the latest lameness from the Wachowski brothers; Elsa & Fred, aimed at old people, a Robert Pattinson vehicle, Orange is the New Black, The Butler and piles of other stuff I would never watch if you tied me to a chair).

Frankly, I don't believe they have a recommendation engine anymore: I think they just suggest whatever is cheapest for them to play.

Re: Netflix and the Napoleon Dynamite Problem (2008)

#43

Post by jblow » Wed, Jul 22, 2015, 6:56 PM UTC

cwbrandsma wrote: post #35

Pxtl wrote: post #30

Earlier quoted context omitted.

Seriously, Netflix' UI is a complete trainwreck. The searching is painful too. I still cannot find a way to say "give me all your View Askew movies" or "all your DC Animated movies". The search is woefully primitive and there is functionally no UI for categories... can't I just pick multiple category tags? Or see all the categories you're using under the hood?

I agree there are some problems with the ui, but at the same time the number of devices they design for is quite large (desktop web, mobile phone, set-top boxes, Apple TV, Google TV, Amazon TV, etc). Creating a UI that is consistent between all of them must be a bit daunting.

So don't do that. Make a few UIs. They are a big company, they can do that if they actually have anyone who knows how to program.

Re: Netflix and the Napoleon Dynamite Problem (2008)

#44

Post by VLM » Wed, Jul 22, 2015, 6:59 PM UTC

A controversial observation might be they maximize profit by having the most people sign up and then stream the least possible, correct? And they have a semi-monopoly, correct, where theres no real competition? I mean, I have and use Prime but I don't seriously watch videos and I'm in it for the shipping...

Anyway there's this assumption that they'd want to present results that encourage you to watch as much as possible, yet they have a financial incentive to present results that are just barely good enough that you won't cancel and/or sign up for a minor competitor.

This is NOT like a public library card catalog where next years funding depends on increasing circulation counts.

In some ways I think my proposed search result criteria are more difficult than presenting result I'd actually want.

Re: Netflix and the Napoleon Dynamite Problem (2008)

#45

Post by s_kilk » Wed, Jul 22, 2015, 7:01 PM UTC

redthrowaway wrote: post #25

mrspeaker wrote: post #14

"If you use a computerized system based on ratings, you will tend to get very relevant but safe answers"... I call this my "Radiohead metric": for a given music recommendation system, how long do I have to use it before it tries to recommend me Radiohead. One of the problems I find in music algorithms is they can't know how I've changed - suggesting music based on what I play or input - but that's stuff I liked in mi…

Applying weight to each rating based on its age seems like such an obvious fix that someone must have tried it and realized that it makes things worse.

Maybe, but I'd think you'd have to take the age weight with a pinch of salt.

The are 14 year olds who are obsessed with jazz and blues, while some 70 year olds only listen to hardcore punk and deathgrind

Re: Netflix and the Napoleon Dynamite Problem (2008)

#46

Post by andreasvc » Wed, Jul 22, 2015, 7:06 PM UTC

stillsut wrote: post #9

quotedmycode wrote: post #2

From what I understand, they never implemented any of the algorithms, because their existing one was good enough, and the premise has changed. It used to be, you'd make suggestions so people could get DVDs sent to them, and you'd know they would enjoy it. Now with streaming, the cost to send something you don't like is cheap. So the recommendation engine just has to come up with something you'd try to watch and perha…

There was a parsimonious model that got to 8% improvement. We know this as it was posted publicly by "Simon Funk" [0] after he decided to give up halfway through the competition. This simple but more accurate model would have been a useful foundation for Netflix to build its own internal models on, and highlight the flaws in the assumption of the Cinermark algo. [0] http://sifter.org/~simon/journal/20061211.html

That looks interesting. I wonder if nowadays that result could be replicated with readily available algorithms from scikit-learn.

Re: Netflix and the Napoleon Dynamite Problem (2008)

#47

Post by lanaius » Wed, Jul 22, 2015, 7:10 PM UTC

kelseydh wrote: post #36

quotedmycode wrote: post #2

From what I understand, they never implemented any of the algorithms, because their existing one was good enough, and the premise has changed. It used to be, you'd make suggestions so people could get DVDs sent to them, and you'd know they would enjoy it. Now with streaming, the cost to send something you don't like is cheap. So the recommendation engine just has to come up with something you'd try to watch and perha…

A lot of Netflix's success comes from finding great, unknown movies and allowing you to discover them. They're great for Netflix because they're cheap for them and provide high value. When Netflix users only decide what movies to watch based only on the titles they recognize from marketing campaigns, their disappointment with Netflix's selection will be higher. Their optimal strategy is to have a recommendation engin…

In Search of General Tso.

Re: Netflix and the Napoleon Dynamite Problem (2008)

#48

Post by pasbesoin » Wed, Jul 22, 2015, 7:20 PM UTC

Netflix doesn't show user comments in the app version of its (streaming) product.

I'm sure there are reasons for this. Nonetheless, it continues to strike me that, in that context, their own, internal information on product offerings is restricted to a two sentence blurb and an unqualified single five start rating metric.

Re: Netflix and the Napoleon Dynamite Problem (2008)

#49

Post by cortesoft » Wed, Jul 22, 2015, 7:22 PM UTC

redthrowaway wrote: post #25

mrspeaker wrote: post #14

"If you use a computerized system based on ratings, you will tend to get very relevant but safe answers"... I call this my "Radiohead metric": for a given music recommendation system, how long do I have to use it before it tries to recommend me Radiohead. One of the problems I find in music algorithms is they can't know how I've changed - suggesting music based on what I play or input - but that's stuff I liked in mi…

Applying weight to each rating based on its age seems like such an obvious fix that someone must have tried it and realized that it makes things worse.

It also might have to do with the time span of the data set... If they only have snapshot data, it would be hard to do anything with temporal information.

Re: Netflix and the Napoleon Dynamite Problem (2008)

#50

Post by saalweachter » Wed, Jul 22, 2015, 7:27 PM UTC

davisclark wrote: post #21

pcmaffey wrote: post #20

Earlier quoted context omitted.

The point of any system should never be to choose for a person, but rather, to improve the quality of available choices.

In that case, aren't you, in effect, choosing for me the set of choices available to me for selection, based upon a preconceived notion of "quality"? The Underground Man goes on to say, “I agree that two times two makes four is an excellent thing; but if we are dispensing praise, then two times two makes five is sometimes a most charming little thing as well.” Would two times two makes five be included in a set of hi…

And Rush said, "If you choose not to decide, you still have made a choice.".

There is no alternative to curation. If you display all the possible options, there is still the order -- humans, being humans, will not thoroughly read and equally consider even a list of a thousand options, so the order will affect their choice. Choosing to display the results in random order, or in alphabetical order, or in chronological order, is still a decision which affects the choice of the user.

Curation cannot be avoided, as long as we are men.

I am reminded of Stanislaw Lem's Demon of the Second Kind.