Topic modelling and emotion analysis of the tweets of British and American politicians on the topic of war in Ukraine

Abstract

This paper focuses on the content and emotive features of four politicians' posts that were published on their official Twitter accounts during the three-month period of the russian invasion of Ukraine. We selected two British politicians – Boris Johnson, the Prime Minister of the UK, and Yvette Cooper, the Labour MP and Shadow Home Secretary of the State for the Home Department – as well as two American politicians, President of the USA Joe Biden and Republican senator Marco Rubio. In the first phase, we constructed a dataset containing the tweets of the four politicians, which were selected with regard to the topic of war in Ukraine. To be considered approved, the tweets were supposed to contain such words as Ukraine, russia, war, putin, invasion, spotted in one context. In the second phase, we identified the most frequent lexical tokens used by the politicians to inform the world community about the war in Ukraine. For this purpose, we used Voyant Tools, a web-based application for text analysis. These tokens were divided into three groups according to the level of their frequency into most frequent, second most frequent and third most frequent lexical tokens. Additionally, we measured the distribution of the most frequent lexical tokens across the three-month time span to explore how their frequency fluctuated over the study period. In the third phase, we analysed the context of the identified lexical tokens, thereby outlining the subject of the tweets. To do this, we extracted collocations using the Natural Language Toolkit (NLTK) library. During the final phase of the research, we performed topic modelling using the Gibbs Sampling algorithm for the Dirichlet Multinomial Mixture model (GSDMM) and emotion analysis using the NRC Lexicon library.