A unique characteristic of text data is the fact that the placement of words in a text sequence has some meaning. Recurrent neural networks (RNNs) are well suited to work with data involving such sequences. Recurrent networks allow output from the previous step to be passed as input to the following step. This process of feeding prior information at a step allows recurrent networks to have memory, which is very useful for dealing with data involving sequences. The name recurrent in RNN also comes from the fact that the output at a step depends on information from the previous step.

RNNs can be used to develop a sentiment classification model where the text data could be movie reviews, tweets, product reviews, and so on. Developing such a sentiment classification model will also need the labels that will be used for training the network. We go over steps for developing a recurrent neural network model for sentiment classification using R in Chapter 10, Text Classification Using Recurrent Neural Networks.