A Deep Learning based hybrid model for improving accuracy of sentiment analysis

Abstract

Text Sentiment analysis has been of great importance over the last few years. It is being widely used to determine a person’s feelings, opinions, and emotions on any topic or for someone. In recent years CNN and LSTM have been widely used to develop such models. CNN has shown that it can effectively extract local information between consecutive words, but it lacks in extracting contextual semantic information between words. Although LSTM is able to extract some contextual information, it lacks CNN in extracting local information. To counter such problems, we used the attention mechanism in our multichannel CNN with a Bidirectional LSTM model to give attention to those parts of a sentence which have a major influence in determining the sentiment of that sentence. Experimental results have shown that our MultiChannel CNN model with Bidirectional LSTM and Attention mechanism has achieved an accuracy of 93.61% which has outperformed the traditional LSTM-CNN, CNN models, and many machine learning algorithms.