Implementing text classification

In this section, we’ll use LSTM to implement a sentiment analysis example over the Large Movie Review Dataset (IMDb, http://ai.stanford.edu/~amaas/data/sentiment/), which consists of 25,000 training and 25,000 testing reviews of popular movies. Each review has a binary label that indicates whether it is positive or negative. This type of problem is an example of a many-to-one relationship, which we defined in the Recurrent neural networks (RNNs) section.

The sentiment analysis model is displayed in the following diagram:

Figure 6.15 – Sentiment analysis with word embeddings and LSTM

Let’s describe the model components (these are valid for any text classification algorithm):

1. Each word of the sequence is replaced with its embedding vector. These embeddings can be produced with word2vec.
2. The word embedding is fed as input to the LSTM cell.
3. The cell output, , serves as input to an FC...