Sentiment Analysis by Novel Hybrid Method BE-CNN using Convolutional Neural Network and BERT

Abstract

© 2020, World Academy of Research in Science and Engineering. All rights reserved. Sentiment analysis also well known as Opinion Mining is one of the important task of Natural Language processing for analyzing the mood of Customers. With the advancement in Internet and Big Data there is terabytes of data available from various Social Media platforms. Numerous models are available for text classification using deep learning techniques. Influenced by the favorable outcome of the deep learning models, we developed a method by using CNN (Convolutional Neural Network) and BERT (Bidirectional Encoder Representation from Transformers). This paper discusses about our novel method of sentiment analysis entitled as BE-CNN that uses 3 layers of Convolutional layer architecture for analysis of corpus collected from twitter related to “#Demonetization”. We demonstrated by our experiments on corpus of 17000 tweets and the results are very encouraging showing that method BE-CNN, proposed by us outperforms other existing approaches of deep learning. We achieved the accuracy of 97% in comparison to other model and hence we can conclude that our method is better than other existing models.