Large-scale and long-term characterization of political communications on social media

• Autor(a) principal: Lucas Santos de Oliveira
• Data de Publicação: 2021
• Tipo de documento: Tese
• Idioma: eng
• Título da fonte: Repositório Institucional da UFMG
• Texto Completo: http://hdl.handle.net/1843/36669
• topic: Political; Social media; Communication; Characterization; Computação – Teses; Mídias sociais – Teses; Redes sociais on-line – Teses; Redes sociais on-line – Aspectos políticos – Teses

Social media play an important role in shaping political discourse, creating a public sphere that enables discussions, debates, and deliberations. Aware of this importance, politicians use social media for self-promotion and as a means of influencing people and votes. As an example of this assertion, in 2018, Brazilians democratically elected for president the far-right candidate Jair Bolsonaro. One of the most surprising feats of this outcome is that his party, PSL, had almost no television time. His victory was only possible because of his supporters’ engagement and activism on social media platforms, such as on Twitter, Facebook, and WhatsApp. In this context, politicians need to decide how to communicate with their voters to build their reputations. While some politicians only share professional communi- cations about their political agenda and activities, others prefer a more non-political and informal approach, sharing communications about the most varied subjects, such as religion, sports, and their families. Others, however, misuse platforms by spreading political messages that violate policies and circumvent electoral laws. Aware of these problems, we propose a supervised machine learning classifier that labels all textual messages from different social media platforms as political and non- political. The classifier runs on a large scale and it is robust to concept drifts over time, requiring few new labeled messages each year. From the classified messages, we were able to characterize the communication of politicians over time and identified new findings: (i) Brazilian congresspeople changed their communication behavior over time; (ii) concept drifts occurred during important events of Brazilian politics; (iii) the explosive rise of the right seen just before the 2018 elections; (iv) a broader and more evenly distributed right-wing participation than the left-wing, and, finally, (v) the increase of public engagement over time.