Analysis of COVID-19 5G Conspiracy Theory Tweets Using SentenceBERT Embedding

Twitter is a popular major social media platform with a central role in the distribution of information, and as such a fertile land for the growth of conspiracy theories in different subjects, with COVID-19 conspiracies among them. In this research, we collected a dataset of 331,448 tweets related to the COVID-19 5G conspiracy theory. We present a workflow to collect, classify, and analyze conspiracy related tweets as supporting or opposing the conspiracy theory. We hand labeled 4,291 tweets and trained a classifier using a novel approach containing two sets of features: a set of sentence embeddings produced by Covid-Twitter-BERT and Sentence-BERT, and a set of external features. We used five different classifiers and ensemble learning to combine them. We classified the dataset and analyzed the classified dataset to conclude that opponents of the conspiracy dominate the conversation on Twitter.