Text clustering and topic modeling on Covid-19 vaccine tweets using machine learning, natural language processing, and deep learning

In late December 2019, the unique coronavirus disease (COVID-19) emerged, causing a tremendous loss of life throughout the globe and posing a previously unheard-of challenge to public health, education, social life, global economies, and the workplace. To end the COVID-19 pandemic, equitable access to safe and effective vaccinations is essential. The best approaches to quickly gain an understanding of COVID-19 text data presented in the literature are those that use unsupervised learning. The goal of this research thesis is to use text clustering and topic modeling to analyze coronavirus vaccine tweets. Using machine learning and deep learning methods and techniques, it investigates the optimal number of topics and clusters prevalent in the coronavirus (COVID-19) vaccine corpus. The study also looks into various insights that can be extracted from tweets through exploratory data analysis, using word embeddings to improve the accuracy of the proposed models, evaluate unsupervised learning methods, and gain other insights. Text clustering was performed using machine learning clustering techniques and algorithms like k-means and HDBSCAN, deep learning-based clustering methods, and dimensionality reduction algorithms such as PCA, LDA, t-SNE, and UMAP, while topic modeling algorithms such as LDA, GSDMM, and TopicBERT/ BERTopic were used to obtain relevant topics from the coronavirus vaccine corpora. The findings of this study demonstrated that GSDMM and BERTopic produced significant topics from the COVID-19 corpus, while deep learning clustering methods outperformed their machine learning counterparts in text clustering. K-means performed superior clustering based on multiple assessment criteria, but HDBSCAN performed better clustering based on features learned.