Effectiveness Assessment of Different Deep Learning Architectures for Sentiment Analysis on Twitter During COVID-19

People are sharing their opinions, feelings, and attitudes about many aspects of the COVID-19 pandemic on Twitter, where social media activity has surged. In order to find the optimum approach, this study compares different deep learning architectures for sentiment analysis of COVID-19-related tweets. This article makes use of a large dataset of tweets concerning the COVID-19 pandemic, encompassing a wide range of opinions, from neutral to unfavorable. Thorough preprocessing of the dataset is carried out, including tokenization, removal of stop words, and application of methods such as word embedding to store semantic information in compact vector representations of words. Examined in this comparative study were three deep learning architectures that saw extensive use: RNNs, CNNs, and transformer-based models. Accuracy, precision, recall, and F1 score are some of the appropriate performance metrics that are used to train and evaluate each architecture. When it comes to conducting sentiment analysis on tweets pertaining to COVID-19, our experimental results provide useful insights into the efficiency and effectiveness of different deep learning architectures. Although all deep learning architectures demonstrate sentiment analysis capability, the study found that some designs offered better accuracy and sentiment prediction capabilities than others. Transformer-based models, such as the famous BERT architecture, perform quite well, particularly when it comes to understanding the complex sentiment dynamics in tweets and absorbing contextual information.