I would have to review that. But there's a problem with speculating
what Netflix does. As Barry would tell you, having worked for tech
companies, it is often not *what* they are actually doing. But any
programming just glancing at the article would as expected it would
speculate that is basically data mining or maybe more correctly "data
scraping."
Netflix will post blog articles about their techniques. And I get to
argue with people who claim they are still using these techniques some 3
or 4 years after the blog article has been posted. Worse yet I get to
argue with geeks about Netflix encoding methods. One ongoing argument
is that my Sony BD is too old to use "adaptive streaming." Mostly
Netflix does not have one file for each stream. Instead streams are
broken up into each small files for MPEG-4 and those may be for HD as
small as 1.2 MB each. They will have several of thees kinds of stream
files at different resolution ranging from 240p to 1080p. If they
detect some congestion between you and their server they will drop down
to a lower resolution stream just to catch up. In fact if you watch a
movie on Netflix they often will use a 240p stream just to start for
about 30 seconds. That's because at the start of most movies is a bunch
of studio and distributor logos so it doesn't matter. In the meantime
they're able to buffer up to a minute of HD while your player shows those.
And that's for ONE kind of steam. Netflix admits they may have as many
as 120 different stream packages for each show. That's to handle the
wide range of devices that Netflix can be viewed on. And it gets worse
than that. The MPEG-4 example just makes it a little easier to
understand. But the files don't have to be broken up into small files
on the server. Silverlight which Netflix uses doesn't necessarily do
that as I've played with Microsoft Expression which can be downloaded
for free and will do up to 10 minutes of Silverlight encoding at
different resolutions. Those are in VC-1 format though the paid pro
version can do MPEG-4. Each resolution has only one file and an index
manifest for seeking.
Favored probably by the industry is the emerging HTML5 codec that Google
offers... for free. This is Webm which has the VP8 or VP9 codec from the
company Google bought. Those are like Silverlight singe files for each
resolution, video only. The audio file is separate. All these have a
meta data or manifest that allows them to quickly seek the segment to
send out. Chromecast is most likely using these for newer films.
Where I get into an argument with the techies is that I have always seen
varying resolution on my supposedly non-adaptive (DASH) supported BD
player. I have even told them that the Netflix app got updated at least
once. I also have done projects using Sony's SDKs and know how they
handle MPEG-4 files. The player is probably about the same as on their
devices I have programmed. BTW, none of the people I am arguing with
have EVER done that nor even shot, edited video let alone created video
players (I've had to create two of those).
The reason Netflix won't tell you *what* they are actually doing is tech
is VERY dynamic. So what they did last year might be completely redone
this year. They keep upgrading their technology. They need to remain
competitive. And when you work for a tech company you are bound by NDAs
which severely limit what you can say. So we don't exactly have Netflix
employees (who are probably lurking and laughing) drop in to post.
So reverse engineering their suggestions technology might have led to
gaffaws and grins down in Los Gatos.
On 02/07/2014 10:05 AM, authfri...@yahoo.com wrote:
Ooh, sorry for the link-lack.
So what did you think of the Perry Mason phenomenon?
They've always been a bit laughable for me because they are
recommending films I've already seen. What Netflix can't do if know
that you've seen via other sources. I rent occasionally at Redbox.
These are often films that Netflix won't be getting for awhile if at
all. For instance, Netflix rarely gets any Universal movies (though
that may change now that Comcast owns Universal and I'll explain
later). So Universal's big films if I am interested in one at all
I'll just rent at Redbox on Bluray. I also rent some indie films at
Redbox again dependent on the distributor or if it is a title I want
to see right away rather than wait maybe a month for it to arrive on
Netflix.
One movie Netflix keeps recommending is "Assault on Wall Street" by
Uwe Boll. Thing is it was definitely a title I was interested in
having seen the trailer posted on conspiracy sites. So when it hit
Redbox I immediately rented it on release day. I even recommended the
film here. About two weeks later it was available to watch on
Netflix. But what was additionally interesting was Boll's commentary
which included why he made the film but also how little money
filmmakers get anymore for their films including how little companies
like Showtime will pay for a film. BTW, I watched Boll's remake of the
1950s film (with Frank Sinatra) "Suddenly" last night on Netflix.
It's well worth a watch and Ray Liotta stars and Dominic Purcell who
played the lead in "Assault on Wall Street" has a supporting role.
BTW, I have 11 Redbox kiosks within 2 miles so I can usually find the
film I want at one. Because of streaming demand on Netflix I often
leave those rentals for the weekend where it might be difficult to get
a decent HD stream from Netflix.
Another place if I am really impatient to see a film is to rent it on
VUDU or Amazon. Both will have films before they are available in
theaters or while in theaters. These aren't the Hollywood
blockbusters but mostly indie films that some distributors like Mark
Cuban's Magnolia Films have taken to releasing online due even prior
to theatrical release due to the dwindling number of art houses. You
pay more for these, though often just the price of a nighttime theater
ticket (no senior discount either).
A bonus with most discs is that they have extras including
commentaries. That is except for the big studios like Universal who
now put "rental" discs at Redbox which have the film only. However
even Universal, now owned by Comcast, might sell Redbox full featured
discs if it is a indie film. Same with Fox. Seldom if ever from
Warner Brothers. Comcast may be adopting what I think is the smarter
business model that Sony Pictures and Lionsgate uses and that is to
get your content in as many channels as possible to maximize return.
So thing is, being a film and TV buff, I am often ahead of the game as
far as knowing what is coming and what I want to see. Also Netflix
supports third party sites. Unlike some other companies they saw the
value of third party fan and review sites as free advertising.
However they did pull the "coming soon" data as distributors started
complaining. Obviously if I knew that a film was going to be on
Netflix two weeks after it arrived at Redbox I would wait the two weeks.
As for the article you recommended I politely thanked you for thinking
of me. But that article came out over a month ago. Not only that you
didn't provide a link. I recall looking at it and the discussion on
the Netflix discussion section of a forum I was on. There we had long
been discussing Netflix's "suggestions" algorithms. And yeah, that
article reveals just what I would have done if I worked at nearby
Netflix (about 60 or 70 miles away as the crow flies) as that project
lead. It is just a form of data mining not that far removed from
having a word processor catalog stuff for a document index. Here's
the missing link:
http://www.theatlantic.com/technology/archive/2014/01/how-netflix-reverse-engineered-hollywood/282679/
On 02/07/2014 04:29 AM, authfriend@... <mailto:authfriend@...> wrote:
*I never pay any attention to them. Why do you ask?*
<< So what do you think about Netflix's recommendations of shows for you?
Well, let's take a look at what you /actually/ said:
"I looked at some articles on this a couple weeks ago. However
their suggestions are about as relevant as what Google or Amazon
recommends..."
Which, you know, makes it appear that "this" refers to the relevance
of Netflix's suggestions. And that (as I said) is not what the
article I recommended is about. Nor is it about what you go on to
talk about in your next paragraph:
You seemed to have missed that I said this information is a couple
weeks old. You are also forgetting that I'm a programmer and was
well aware of heuristic procedures being used to determine tastes.
In fact I am on home theater forums where this is discussed quite a
bit including your aforementioned article. And being in the
entertainment industry know how we "engineer" products to appeal to
tastes.
It seems that you do not, in fact, know what this particular article
is about. Nothing to do with determining or appealing to tastes, you
see.
But heaven forbid you actually read the article and find out.
Oh, and what did the folks on your home theater forums have to say
about Perry Mason? >>
On 02/06/2014 09:33 AM, authfriend@... <mailto:authfriend@...>
wrote:
*The relevance of Netflix's suggestions is NOT NOT NOT what this
article is about, Bhairitu. If you actually have a look at it, I'm
pretty sure you'll be intrigued.*
Thanks. I looked at some articles on this a couple weeks ago.
However their suggestions are about as relevant as what Google or
Amazon recommends because of the way I use Netflix. For instance I
only watched "Atlas Shrugged II" for reference and gave it only 1
star (you can't give no stars) so they post a message after such a
rating that they have no recommendations based on that rating. The
movie itself is quite laughable.
On 02/05/2014 09:37 PM, authfriend@... <mailto:authfriend@...>
wrote:
How Netflix Reverse Engineered Hollywood
To understand how people look for movies, the video service
created 76,897 micro-genres. We took the genre descriptions, broke
them down to their key words, … and built our own new-genre generator.
This article from The Atlantic by Alexis Madrigal is a whole lot
more fascinating than it sounds. Especially the Perry Mason Ghost
in the Machine, which emerges toward the end. The "new-genre
generator" is the least of it.