Long short-term memory (LSTM) networks are a special type of recurrent neural network. LSTM networks are useful when data regarding the sequence of words or integers has long-term dependencies. For example, two words that are important for correctly classifying sentiment contained in a movie review may be separated by many words in a long sentence. A sentiment classification model using a regular RNN will have difficulty capturing such long-term dependency between words. A regular RNN is useful when dependency between words or integers in a sequence is immediate or when two important words are next to each other. 

Apart from sentiment classification, the application of LSTM networks can also be useful for speech recognition, language translation, anomaly detection, time series forecasting, answering questions, and so on. An application of an LSTM network for movie review sentiment classification is illustrated in Chapter 11, Text Classification Using Long Short-Term Memory Network.