How Did Europe's Press Cover Covid-19 Vaccination News? A Five-Country Analysis

Abstract

Understanding how high-quality newspapers present and discuss major news plays a role towards tackling disinformation, as it contributes to the characterization of the full ecosystem in which information circulates. In this paper, we present an analysis of how the European press treated the Covid-19 vaccination issue in 2020-2021. We first collected a dataset of over 50,000 online articles published by 19 newspapers from five European countries over 22 months. Then, we performed analyses on headlines and full articles with natural language processing tools, including named entity recognition, topic modeling, and sentiment analysis, to identify main actors, subtopics, and tone, and to compare trends across countries. The results show several consistencies across countries and subtopics (e.g. a prevalence of neutral tone and relatively more negative sentiment for non-neutral articles, with few exceptions like the case of vaccine brands), but also differences (e.g., distinctly high negative-to-positive ratios for the no-vax subtopic.) Overall, our work provides a point of comparison to other news sources on a topic where disinformation and misinformation have resulted in increased risks and negative outcomes for people's health. CCS CONCEPTS• Computing methodologies → Natural language processing; • Information systems → Data mining.

Graph-Based Interpretability for Fake News Detection through Topic- and Propagation-Aware Visualization

“…Although our framework is compatible with various text-clustering methods, for this investigation, we chose BERTopic, which is a state-of-the-art text-clustering technique for topic modeling, because of its robustness and accuracy. In recent years, BERTopic has been employed as a method for various data analyses of news and social media posts, such as the topic analysis of profiles of users who mentioned QAnon [37], topic analysis as a comparison material for the veracity of news sources [38], and temporal changes in posts with topics of vaccine skepticism and denial [39]. In the following section, we review text clustering via BERTopic and explain the functionalities of our visualization method.…”

Examining European Press Coverage of the Covid-19 No-Vax Movement: An NLP Framework

“…Our starting point has been a previous work [3], where we created a dataset of more than 50,000 articles on Covid-19 vaccination with articles from Italy (2 newspapers), France (6 newspapers), Spain (6 newspapers), Switzerland (3 newspapers) and the United Kingdom (2 newspapers), with all the content translated to English.…”

“…Using the results of the sentiment analysis carried out at article level, sentence level, and headline level in our previous work [3], we were able to compute the majority sentiment for each political orientation. So that each headline has a sentiment.…”

Framing the News: From Human Perception to Large Language Model Inferences

“…We used part of the European Covid-19 News dataset collected in our recent work [3]. This dataset contains 51320 articles on Covid-19 vaccination from 19 newspapers from 5 different countries: Italy, France, Spain, Switzerland and UK.…”

“…In Figure 5, we compare the sentiment per frame and per country, to understand if there were any major differences. The sentiment analysis labels were obtained using BERT-sent from the Hugging Face package [47], used in our previous work (please refer to our original analysis in [3] for details.) We normalized the results between 0 and 1 to compare frames between countries.…”