Word embeddings

The order of words in a text matters. Therefore, we can expect higher performance if we do not just look at texts in aggregate but see them as a sequence. This section makes use of a lot of the techniques discussed in the previous chapter; however, here we're going to add a critical ingredient, word vectors.

Words and word tokens are categorical features. As such, we cannot directly feed them into a neural network. Previously, we have dealt with categorical data by turning it into one-hot encoded vectors. Yet for words, this is impractical. Since our vocabulary is 10,000 words, each vector would contain 10,000 numbers that are all zeros except for one. This is highly inefficient, so instead, we will use an embedding.

In practice, embeddings work like a lookup table. For each token, they store a vector. When the token is given to the embedding layer, it returns the vector for that token and passes it through the neural network. As the network trains, the embeddings get optimized...