Evaluting BERT and Neural Network Approaches for Tweet Sentiment Analysis

Abstract

Deep Learning is being explored as a prominent tool in machine learning in the current era by researchers to a great extent. A lot of work being done in this area is opening avenues for consumers. Utilization and growth of Artificial Intelligence have leveraged the power of deep learning for its applicability. Since its inception, a lot of algorithms have been proposed and implemented in the area of deep learning. In this paper, BERT (Bidirectional Encoder Representations from Transformers) a model of Google, LSTM, DistilBERT and CNN models have been discussed. Deep learning models have a major challenge of low performance, which restricts their use in real-time applications. In this work, BERT, LSTM, DistilBERT and CNN have been implemented separately using the same dataset of Twitter to perform sentiment analysis and compared their performance and accuracies as a first step to make a new model that will not only provide good accuracy but will also have better accuracy in future.