Three Essays on Political Polarisation: A Computational Analysis of the Italian Political Discourse on Twitter (2014-2019)

最近的研究表明，民粹主义两极分化的社交媒体传播导致了公众对民主制度和政客的负面态度。然而，政治学家未能全面理解此类沟通策略为何以及如何对网络行为产生负面影响。本论文旨在通过调查民粹主义者在社交媒体上使用情感极化语言及其对在线政治行为的影响，衡量西欧人对帖子的反应，来弥补文献中的这一空白。特别是，这项研究对意大利、德国、法国和英国的主要民粹主义政党的社交媒体传播进行了计算分析。它结合了尖端的监督机器学习、深度学习、计量经济学和网络抓取方法，研究了2013至2020年间Twitter和Facebook上的民粹主义交流。这篇博士论文具有重要的理论、规范和方法论意义。从理论上讲，它在社交媒体上引发了关于政治经济学、心理学和计算社会科学的持续辩论。特别是，每一篇论文的理论贡献都被结合成一个全面的情感极化理论，并通过假设进行检验，旨在把握政治经济学、同质性和信息加工与AP之间的关系。规范地说，这篇论文为被分析国家的在线行为政策做出了贡献，也为全球科技公司更好地识别和解决在线两极分化问题的研究做出了贡献。最后，本文综合运用了前沿的监督机器学习、深度学习、计量经济学和网络抓取等方法。这样做可以让这项研究通过将计算方法应用于社交媒体上的政治在线行为背景，对AP的广泛影响做出经验性贡献。总而言之，本研究项目有可能对政治话语中两极分化语言的使用及其在国家和时间上的影响提供强有力的理解。这一结果可能会揭示网络两极分化如何影响反应，以及它对西欧和其他地区的网络行为的广泛影响。

Recent studies show that populist polarised social media communication leads to the public's negative attitudes towards democratic institutions and politicians. However, political scientists have failed to produce a comprehensive understanding of why and how such communication strategies negatively affect online behaviour. This dissertation aims to contribute to this gap in the literature by investigating the populist use of affective polarised language on social media, and its effects on online political behaviour, measuring responsiveness to posts in Western Europe. In particular, this research provides a computational analysis of the major populist parties' social media communication in Italy, Germany, France and the UK. With a combination of cutting- edge supervised machine learning, deep learning, econometric and web-scraping methods, it studies populist communication across Twitter and Facebook between 2013 and 2020.This doctoral dissertation has significant theoretical, normative, and methodological implications. Theoretically, it contributes to ongoing debates regarding political economy, psychology and computational social sciences on social media. In particular, the theoretical contributions of each paper are combined into a comprehensive theory of affective polarisation, and tested through hypotheses, which aim to grasp the relationship between political economy, homophily and information processing and AP. Normatively, this dissertation contributes to online behaviour policies in the countries under analysis, and to global tech companies' research to better identify and address online polarisation issues. Finally, this dissertation utilises a combination of cutting- edge supervised machine learning, deep learning, econometrics and web-scraping methods. Doing so allows this research to contribute empirically to the broad effects of AP by applying computational methods to the context of political online behaviour on social media. In conclusion, this research project has the potential to provide a strong understanding of the use of polarised language in political discourse and its effects across-country and across- time. The results may shed light on how online polarisation affects responsiveness, and its broad consequences on online behaviour in Western Europe and beyond.