What makes WiFi faster at home than at a coffee shop? How does Google order its search results from the trillions of webpages on the Internet? Why does Verizon charge $15 for every GB of data we use? Is it really true that we are connected in six social steps or less? These are just a few of the many intriguing questions we can ask about the social and technical networks that form integral parts of our daily lives. This course is about exploring the answers, using a language that anyone can understand. We will focus on fundamental principles like “sharing is hard”, “crowds are wise”, and “network of networks” that have guided the design and sustainability of today’s networks, and summarize the theories behind everything from the social connections we make on platforms like Facebook to the technology upon which these websites run. Unlike other networking courses, the mathematics included here are no more complicated than adding and multiplying numbers. While mathematical details are necessary to fully specify the algorithms and systems we investigate, they are not required to understand the main ideas. We use illustrations, analogies, and anecdotes about networks as pedagogical tools in lieu of detailed equations. Please note that per Princeton University policy, no certificates, credentials or reports are awarded in connection with this course.
Viral style and video recommendation
Loading...
来自 Princeton University 的课程
Networks Illustrated: Principles without Calculus
39 个评分
从本节课中
Viral Videos on YouTube
What does it take for a video to become "viral" on YouTube? In this lesson, we will take a look at some of the key factors and models that have been used to explain this phenomenon. At the core is the notion of information cascade in a network, which is the counterpart to the wisdom of crowds.
与讲师见面
Christopher BrintonLecturer
Electrical Engineering
Mung ChiangProfessor
Electrical Engineering
Looking at the concept of viral, note
that YouTube offers some interesting tools, like
YouTube Insight, YouTube Statistics, Analytics that highlight
the overall viewing behavior, on individual videos.
Using these analytics, we can look at the
trends behind those videos that have gone viral before.
So, the pinnacle of what we say viral is today, is the video Gangnam Style, which I
did not mean to rhyme but it just turned out to rhyme that way, viral and style.
And this is a four minute music video by Psy, that you probably all
know, which is the first and only video to have reached over one billion views.
So it, it's the only video that has one
billion views, and it did this in just five months.
So, what we can, we can see here by looking just at the
analytics or statistics behind the video which if you go to the YouTube page
for Gangnam Style you can see a more updated one through the present day
and you can, you can check out the total view counts and so forth.
We see you know, it's uploaded and July 15th of 2012 and
within five months roughly it had broken that one billion view count.
And by April 2013, it actually hit, about 1.5 billion views.
So, that's up here now.
We see that it's over that total
1.5 billion, which, which is really remarkable.
And this section highlights two key things that are present
in viral videos, which we'll, we'll talk about and formalize
more soon, is that it needs to have a high
total view count and also a short path to getting there.
So, it can take a long time to get to that high total.
This really brings us to the driving question behind this lecture,
which is how do, how do you viralize a YouTube video?
And in fact, there's been a lot of research on how to make a video go viral.
The most logical starting point to that is to look at forming paths by
which a, a user could be led to YouTube video in the first place.
So let's, let's look at ways in which people
can get to YouTube videos, because total views is what's
going to determine whether or not a video goes viral
or not, and how quickly those views are coming in.
So the first path could be through a web search.
Like from Google.
So when you Google something a YouTube video can come up.
Also YouTube search itself, obviously.
So YouTube search when you, when you're on YouTube, and
you type in a query for a, a given video.
You know, that video could pop up, and then you can click on it.
Second is referral through a variety of means, a
variety of sites like email, Twitter, Facebook, and so forth.
Third is through subscription to a YouTube channel.
So if you're subscribed to someone's channel, and they upload
a new video you could be recommended to go see it.
And fourth is through browsing through the YouTube recommendation page.
So if this is a, a YouTube page that you're looking at
you know, the, the video when you're not in full screen mode
is up here you know you can play the video, pause the
video so you have some, some video playing up in the screen.
Down here is the the comment section where,
you know, people are writing in different comments.
You know, hating on each other, saying this is the
worst video ever, the best video ever, the fights start,
et cetera, you know it, you know it well and
then you know you have your likes and dislikes up here.
And then, on the right hand side, though, you have
these, this section that says something like, recommended to you.
So recommended to you.
[BLANK_AUDIO]
And there's a whole bunch of videos on the side that are
recommended based upon the current video
that you're watching and also other information.
And note that often, the subscription and recommendations,
so these, these last two points, are going to play
a bigger role in a video's popularity, than
the number of likes and dislikes that it has.
We've looked at ranking and recommendation in other lectures.
So let's take a moment to reflect on how YouTube determines
what the recommendations of other videos based upon the current video.
What YouTube does, they look at what's called the Co-visitation count.
And co-visitation count is how often a pair of
videos are watched together by viewers in some time frame.
So if I have two videos, say A and B, and I have say
100 people, that have watched both videos.
So 100 people have watched both the videos A and B.
In say 24 hours.
So maybe 24 hours.
And the co-visitation count for A and B, I would say is 100.
And so again, the same person watching both videos.
The total number of people that satisfy that.
So Alice, Bob, and so forth haven't watched videos A and B in the last 24
hours, would give the co-visitation count currently of 100 for videos A and B.
So, we can actually define a graph from this.
Right, where the videos are the nodes.
so, here the circles are going to be videos, and the
link weights of number the people that have watched both videos.
So we connect videos if they have some co-visitation count.
Meaning that there's at least one person that's
watched both videos in the last say, 24 hours.
And then the, the link weights are going to be the co-visitation count.
And so, intuitively here, if we're in this black node and we're trying to determine
what the recommendations will be, you know assuming
that the videos are related to one another.
Which we would have to use, to, we'd
have to combine with keyword match to determine exactly.
But let's assume that these four videos here, are all relatively the
same in terms of having the same keywords as this center video here.
Then most likely, we will be recommended first to
this one that has the co-visitation count of 80.
Followed by this one with 40, followed by this
one with 30, followed by this one with 20.
Obviously this graph here is nowhere near the size or scale of YouTube as a whole.
There's going to be, you know, many, many more
connections, and many, many more nodes than here.
But this is just a small example that we can
fit on a page, as we've said many times before.
And so, the combination of this co-visitation count
with keyword matching is going to determine the recommendations.
Based upon the current video.
So note that this is, this is a kind
of a simpler scheme than other ones that we've seen.
Like with page rank we saw how we would,
we would leverage hyperlinks to define a web graph.
And then use that to come up with a ranking scheme.
And here, we're not going to use, we would have to use
what are, the tags of the videos which would tag related videos.
So forth, to determine which point to which.
And those, those are kind of unreliable for YouTube videos.
You know, they're, they're, they're user generated.
And sometimes people put tags in just to try to increase their
total view count and get more people to come and view their video.
Not really indicative of the content even within the video.
And it's also not sophisticated Netflix style where we, we try
to predict what a person's rating would be for a video.
And the reason for that is that viewing [INAUDIBLE] for
these short YouTube clips is really going to be too variable.
And so it's, it's not going to be easy
to determine not going to be easy to factor in
how much of the videos a person watched
based about whether they liked or disliked the video.
Just, just not as clear.
Not as intuitive in that case.
So, YouTube uses instead the co-visitation count, which works quite well,
especially if you think that each of recommendation works well yourself.
[BLANK_AUDIO]
