Last week, you looked
at tokenizing text. Where turn text into
sequences of numbers with a number was the value of a key value pair with
the key being the word. So for example, you
could represent the word TensorFlow
with the value nine, and then replace
every instance of the word with a
nine in a sequence. Using tools and TensorFlow, you are able to
process strings to get indices of all the words in a corpus of strings
and then convert the strings into
matrices of numbers. This is the start of getting sentiment out of your sentences. But right now, it's
still just a string of numbers representing words. So from there, how would
one actually get sentiment? Well, that's
something that can be learned from a corpus of words in much the same way as features were
extracted from images. This process is called embedding, with the idea being
that words and associated words are clustered as vectors in
a multi-dimensional space. Here, I'm showing
an embedding projector with classifications
of movie reviews. This week, you'll learn
how to build that. The reviews are in
two main categories; positive and negative. So together with the labels, TensorFlow was able to build
these embeddings showing a clear clustering
of words that are distinct to both of
these review types. I can actually search for words to see which ones match
a classification. So for example, if I
search for boring, we can see that it lights up in one of the clusters and that associated words were clearly negative such as unwatchable. Similarly, if I search for
a negative word like annoying, I'll find it along
with annoyingly in the cluster that's clearly
the negative reviews. Or if I search for fun, I'll find that fun and
funny are positive, fundamental is neutral, and unfunny is of
course, negative. This week, you'll learn how
to use embeddings and how to build a classifier that
gave that visualization. You're most of the way there
already with the work that you've been doing with
string tokenization. We'll get back to that later
but first let's look at building the IMDB classification that you just visualized.