Abstract

Internet technology has grown to the extent that people create, share ideal, opinion and content on Twitter. Useful information is obtained from Twitter on Coronavirus (COVID-19) for response and management of the pandemic. The classification of Twitter is affected by addressing negation for sentiment analysis and domain dependence for Twitter events or topic. In this paper, ensemble deep learning models are used to learn and train network on real world COVID-19 Twitter data in analysing and provide a broad classification to user’s opinion and events discuss in multiclass and multilabel, respectively. A Textblob lexicon and Natural Language ToolKit (NLTK) are employed for polarity sentiment and event discuss for categorisation and event similarity respectively, this reduce the bias in the network and enhanced models’ performance. A comparative analysis is performed on Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) and C-LSTM with pretrained distributed GloVe embedding. The classification is performed in batches to reduce the complexity and memory usage, however, dropout and early stopping strategies are employed to prevent overfitting with Adaptive Momentum (ADAM). Evaluation of the models are performed using precision, recall, F1 score and accuracy with 90%, however, LSTM shows more accuracy compared to CNN and C-LSTM.