A Comparative Study of Deep Learning Neural Networks in Sentiment Classification from Texts

Nowadays, text data has become the dominating medium for expressing views and opinions on various issues on online platforms. The necessity of mining information and polarity from public opinion is outgrowing. The analysis of sentiment on social networks, such as Twitter, Facebook and E-commerce websites is an effective way of learning about the users' opinions and reviews. In recent years, it has been experimentally shown in some research works that deep learning neural networks are a promising solution to the challenges of Natural Language Processing. This research conducts a comparative study among some neural networks such as LSTM, Bi-LSTM, GRU-CNN and the combination of CNN and RNN. Two datasets from different domains have been used: The IMDB Movie Review (50 k), which is a benchmark dataset and an Amazon Product Review (10 k) dataset. Amazon is also a standard platform for collecting datasets. Analyzing the experimental results, we realize that CNNs can sometimes outperform RNNs. Also the comparative performance may differ in different domains and datasets.