[MUSIC] Let's dig in and talk a little bit
about the capstone application that we're gonna build together. So we're gonna build a recommender
system for products that's intelligent, that combines text, images,
sentiment analysis, and deep learning. Now, before talking about the capstone
project which is really going to be exciting. Let me give you a quick demo of
something that you could build, a recommender system that you could build,
that combines text and image data and uses deep learning at its core. Now let me show you an example of an
intelligent application for recommending products that combines image and text
data and uses deep learning at its core. So let me start off with a story. So my sister's a fashion designer, and
her birthday was coming up this month, and I wanted to get her a gift
that would impress her. I want to get her a dress. But it's really hard for
me to choose a good dress. I don't know much about dresses. So if I do just standard keyword
search and I search for dresses. I am going to end up with something that
looks like this that I am showing you on the right here. And you will see there's a wide range
of dresses and it is very hard for me to describe the kind of dress
that my sister likes and so I have seen her for
example wear some floral dresses so I could of course search that combines
the keywords dress and floral, and looks through the description of the dress
and tries to find some floral dresses. But even within floral dresses, you see that there is a wide
range of floral dresses up there. And each one of them has
a pretty different kind of style. So the question is, how can I describe
what is a good style for my sister? Now, for example,
take this dress over here. It looks a little bit
like what she might like. Those warm sunset colors and rich. So there's some text analyses
that we can do that shows you a little bit about what describes this
dress but really visually looks good. So the question is can we use visual
cues to find other dresses that look kind of like this one? And you do things like this
in your capstone project. When I click on this dress, we're gonna find visually similar dresses
using a technique called deep learning. Then you're gonna learn about
already in this first course and apply it in practice. So when I click on this dress,
what I'm gonna see on the right is a bunch of dresses
that are visually similar to this one. And you will see, for example,
the dresses tend to be floral, they tend to have the similar hues, similar colors, and even then, it's hard
to describe what my sister might like. But I can look at this dress
right in the middle over here, and say wow, this has Interesting pattern,
different colors. Maybe this is what my sister likes. And if I click on it, I'm gonna find
visually similar dresses to this one, so dresses that have more
interesting patterns and multiple colors associated with them. So you'll see as I scroll,
there's variants of these. And then maybe my eye
catches this one over here. Where it looks like something
that my sister might like and I might call her up and say,
hey I think I found a great dress for you. This one. Now let's say that I call my sister and
she looks kinda like this model. I call her up and she says, you know. This looks okay, but
I am going to a cocktail party and why don't you get me a cocktail dress. So, I am trying to think about,
I start over. And I'm going to try to
find her a cocktail dress. Cocktail dresses are ones
that you wear for more formal parties that
look kind of like this. And there's all sorts of colors and
things. But maybe she's interested
in a black cocktail dress. So just play some keyword search like
black cocktail dress, you'll get things like this, but it's hard to describe
what my sister might be interested in. So I might ask her. What you're interested in. She says you know what I'm interested in
is a a cocktail dress that's interesting, with a touch of color. She's a fashion designer. What do I do? How do I describe that? So I look for these and say well, this first few on the left
is kind of interesting. If you look at the keywords associated
with it, you see the keyword jazzy. So maybe she's interested
in a jazzy dress. So what do jazzy dresses look like? So when I click on this keyword,
that I discover from the text, you see other jazzy dresses and
they're kind of interesting but they're not exactly kind
of cocktail dresses. So let me just click on this one and
find a visually similar cocktail dress. Now what you see is a bunch of
cocktail dresses that tend to have a bit more of interesting
patterns associated with them. So they're more interesting, more along
the lines of what she was describing. So if I scroll through this,
say okay, this look pretty cool. Maybe this one in the center, which is
blue, maybe it has some color to it, is what she's looking for. I click on that. I figure out, you know what? It's not formal enough for
a cocktail dress but if I scan down, you'll see that I can find a dress like
this one where it's a cocktail dress, it's formal enough it has a touch of
color to it and it's interesting. And it might be something that she'll
actually be interested in wearing. So here, we've see an intelligent
application that looks at text data, image data, uses deep learning, and does
some really cool and interesting things. So we saw this demo of deep learning for
visual product recommendation. But in your capstone, you do something really even cooler,
much more interesting than this. You're gonna take your
capstone project and you're going to combine
a variety of things. You combine recommenders
with a recommender a system. You're gonna do some text analysis,
sentiment analysis, what people are saying teviews about different products and
extract those reviews and analyze them. You're going to do some computer
vision to visually understand images. You're going to do some deep learning
to really take that computer vision techniques and
make them extremely accurate. And then you're gonna take it and deploy it like a web service like
the one I just showed you, that webpage. You're gonna have your very own
intelligent web service for products which you can interact with,
do some really interesting things and really show to a ton of people how you
created intelligence behind that service. [MUSIC]