Sentiment Classification of Tweets Using Hybrid Transformers-CNN Architecture on the Twitter Platform

Abstract

Twitter, now known as X, is a popular platform used to express opinions on the latest trends, making it a valuable source of data for sentiment analysis research. The huge volume of data makes manual analysis impractical because it requires a long time and human resources, so it is necessary to automate the sentiment classification process through machine learning. Machine learning can be used to classify sentiment on a large scale quickly and accurately by utilising patterns. Machine learning models such as Transformers-CNN show the most superior performance with accuracy reaching 85.71% on test data and 99.90% on training data. The accuracy on the test data was better than other architectures namely LSTM, CNN, BERT, Transformers-LSTM, and LSTM-CNN with accuracies of 84.73%; 82.27%; 77.34%; 85.71%; 84.24% respectively. Transformers-CNN also has a training time of 30.17 minutes which is shorter than Transformers-LSTM, but longer than the other architectures.