Análises de modelos de predição de tweets de jornais e de seus comentaristas sobre as vacinas de Covid-19

Abstract

This work proposes an analysis of tweets written by two newspapers (The Guardian and The Economist), and the comments received (replies), about the term "vaccine" in the year 2020. A sample data collected from Twitter has 1,343 tweets written by two newspapers and 11,784 comments. For the analyses, the following techniques were used: sentiment analysis (Vader), toxicity analysis (Perspective API), unsupervised topic analysis (LDA), and prediction models based on a logistic regression classifier. The Analyzes indicates differences between the newspapers, and The Economist is 40% more positive than The Guardian in their publications on Twitter, this difference was not observed for comments. However, we saw a spike in toxic comments in the weeks of approval for the Pfizer and Moderna vaccines, and also when the first person was vaccinated with the first dose in the UK. The proposed prediction models have 92% accuracy for the newspaper tweets based model and 68.5% accuracy for the comments based model.