Sentiment analysis on risk communication about health: News VS Twitter (based on texts about the COVID-19 pandemic)
Zoya I. Rezanova, Yulia E. Sypchenkova · Vestnik Tomskogo gosudarstvennogo universiteta Filologiya · 2024
The article presents the results of a comparative analysis of sentiment presented in Russian-language news discourse and Twitter discourse while reflecting the dynamics of the pandemic and the accompanying event series. The analysis was performed using NLP methods. The material for the analysis was as follows: texts that in any way reflect the theme of the coronavirus and were published by RIA and TASS agencies in 2020. The volume of materials about Moscow was 3135 texts, and the volume about regions was 1404 texts. To immerse the texts about the coronavirus in the general context of the event radar of this period, the analysis included texts with "incidents", "science", and "culture" topics in the volume of 66,096 texts (31,890 texts reflect events in the regions and 34,206 in Moscow). The sentiment of Twitter texts was analyzed using 178,673 texts. The RuBERT-based blanchefort/rubert-base-cased-sentiment model was used to analyze the sentiment of news texts, and the sismetanin/rubert-rusentitweet model (https://huggingface.co/sismetanin/rubertrusentitweet) was used to analyze Twitter texts. The results of the sentiment analysis were connected with official statistics on the development of the pandemic during this period (https://yandex.cloud/ru/marketplace/products/yandex/coronavirus-dashboardand-data) and information on the chronicle of events (https://ria.ru/20210305/koronavirus-1599707836.html). The analysis showed that news about the coronavirus had a mostly neutral tone, while the tone of the general news flow was negative. No changes in the sentiment were found in Twitter texts: tweets of negative and neutral sentiment were equally distributed almost throughout the whole period under study. What unites the two discourses is the tendency to minimize the number of texts carrying positive sentiments. When analyzing news texts about the coronavirus, we identified several dates on which the negative sentiment was out of the general distribution. The increase in the negative sentiment on certain dates in June and November is connected with the peak of disease incidence in the regions – July and the beginning of the second wave of disease growth in 2020 in November and a new wave of restrictive measures. As the analysis showed, the sentiment of the news flow was influenced by the statistics on the spread of the coronavirus and the government's response to the situation by introducing regulatory measures that affected almost all aspects of social and private life in the country. We also documented days when the level of positive sentiment in news content increased in response to a decrease in disease incidence and reports of vaccine development. In general, the findings are consistent with research conducted on Englishlanguage news articles from COVID-19-affected countries around the world.