Long short-term memory (LSTM) deep neural networks for sentiment classification

Abstract

Recurrent neural networks (RNNs) are useful for text data classification problems. However, when a sequence of words in text data has long-term dependencies, RNNs suffer from “vanishing gradient problem” that makes network training difficult for long sequence of words or integers. Long short-term memory (LSTM) neural networks are a special type of RNNs that help overcome this problem and make them possible to capture long-term dependencies between keywords or integers in a sequence that are separated by a large distance. This chapter provides an application example and illustrates steps for using LSTM deep neural network for movie review sentiment classification. The steps include text data preparation, creating LSTM model, training the model, and then assessing the model performance.