Python Artificial Intelligence Projects for Beginners

skip-gram
, and the center
word is the input and the context words are the outputs:
In this technique, the center word vector is used to predict the context words given that
center word.
Both of these techniques perform well for most situations. They each have minor pros and
cons that will not be important for our use case.
Doc2Vec
We're going to use Word2Vec to detect positive and negative product, restaurant, and
movie reviews. We will do so with a slightly different form of Word2Vec known as
Doc2Vec
. In this case, the input is a document name, such as the filename, and the output is
the sliding window of the words from the document. This time, we will not have a center
word:

---

Applications for Comment Classification
Chapter 3
[ 66 ]
In this case, as a vector that helps us predict the words, from knowing the filename. In fact,
the input is not very important, which in this case is the filename. We just need to keep
track of the words on the right side, and that they all came from the same document. So, all
of those words will be connected to that filename, but the actual content of that filename is
not important. Since we can predict the document's words based on its filename, we can
effectively have a model that knows which words go together in a document. In other
words, that documents usually talk about just one thing, for example, learning that a lot of
different positive words are used in positive reviews and a lot of negative words are used
in negative reviews.
Document vector
After training, we have a new document and we want to find its document vector. We'll use
the word similarities learned during training to construct a vector that will predict the
words in the new document. We will use a dummy filename since the actual name is not
important. What's important is that it's just one name. So, all of these words get connected
together under that one name:
Once we get that new document vector, we can compare it with other document vectors
and find which known document from the past is the most similar, as follows:

---

Applications for Comment Classification
Chapter 3
[ 67 ]
Thus, we can use 
%PD7FD
 to find which documents are most similar to each other. This
will help us detect positive and negative reviews because, ideally, the positive reviews will
have document vectors that are similar to each other and this will be the same for negative
reviews. We expect 
%PD7FD
 to perform better than bag of words because 
%PD7FD
 learns
the words that are used together in the same document, so those words that are similar to
bag of words never actually learned any information about how similar the words are
different.