Sentiment Prediction of IMDb Movie Reviews Using CNN-LSTM Approach

This article describes sentiment classification of movie reviews given by the user using deep neural networks. Long short term memory (LSTM) and convolutional neural network (CNN) are two popular deep neural networks used for sentiment analysis. Sentiment analysis is carried out on internet movie dataset (IMDb) which consist of 50K movie reviews. CNN and LSTM architectures are used individually and later combination of CNN-LSTM architecture is used. Accuracy and loss metrics measures are plotted for each architecture where LSTM architecture outperforms compare to CNN and CNN-LSTM architecture. Accuracy of GRU, CNN, LSTM and CNN-LSTM architecures are 53% 85%, 87% and 85% respectively. Adam optimizer and binary cross entropy is used for loss function. CNN-LSTM model is very good for long term dependency and accuracy is also good. Combination of CNN-LSTM reduces a training time for larger dataset and CNN has a convolutional layer to extract information by a larger piece of text.