Understanding the dynamics of information and misinformation requires systematic evidence about who produces, circulates, endorses, and contests political narratives. Yet social media research frequently relies on content-level indicators that neglect user attributes, suffer from substantial representation and measurement errors, and create serious privacy risks. This paper introduces the Politus Dataset, a large-scale, privacy-preserving political public opinion dataset constructed from approximately 450 million posts collected from X in Turkey.

We are in the era of generative AI, with large language models (LLMs) achieving outstanding results in various tasks. This new technology is here to stay and is already being embedded in our day-to-day lives. However, "with great power comes great responsibility" and researchers have the duty to shed light on all the aspects of genAI. In this talk I will briefly show how LLMs are built and delve into their "good" and "bad" capabilities. I will showcase some of our work leveraging LLMs for multilingual transfer (the "good"), whilst also presenting our work on LLM-generated disinformation (the "bad").

In this talk we discuss how climate misinformation is evolving and the growing role of generative AI in this scenario. In particular, we discuss misinformation strategies through time, to what extent human perception and LLM perspective on climate change are aligned, and the challenge of properly measuring and characterizing AI and its influence. We resume presenting some conclusions as well as current trends and issues.

Climate disinformation erodes public trust in climate science, undermines urgency around extreme weather events, and fuels polarization at exactly the moments when clear information matters most. This is especially relevant here in Rio Grande do Sul, which has faced severe flooding events since 2023–2024 and again in 2026: whenever an extreme weather event unfolds, social media floods with disinformation in near real time, right when accurate information matters most. Yet climate disinformation occupies a distinct space within the broader disinformation landscape: it combines scientific complexity, seasonality, and heavy political charge, and rarely receives dedicated attention at AI-and-disinformation events, which tend to focus on politics or public health. This talk explores how AI — paired with an unlikely ally, football — can help address that gap with ClimaVAR, an AI-based tool built around a simple metaphor: football's VAR (Video Assistant Referee). ClimaVAR combines existing building blocks (classifiers, RAG, and LLMs) into a system accessible enough to be adopted by two real-world communication campaigns during the World Cup 2026. Drawing on this experience, I present what made the approach work in practice, and how it could extend beyond football and climate to other sports and other domains of technical-scientific disinformation.

Computational social science has transformed the study of collective behaviour by making large-scale digital traces available for scientific observation. Yet, as the field has matured, the challenge has shifted from collecting more data to asking better questions of the data we already have. Drawing on a series of studies of Brazilian political communication — including bots, toxicity, cross-platform communities, information cascades, and computational models — this talk reflects on how our understanding evolved as new representations and analytical approaches revealed mechanisms hidden behind aggregate observations. Rather than presenting a sequence of independent studies, the talk argues that many advances in computational social science have come from refining what we observe and how we interpret digital traces. The presentation concludes by discussing the growing limitations of platform-based observations and the complementary role of computational models in explaining social processes, arguing that the future of the field lies in combining observation, representation, and mechanism to better understand increasingly complex digital societies.

In the past and present decades, politics have gotten further alienated from people: The positions of the wealthiest influence political positions more than popular opinion (Gilens and Page, 2014), corporations are now classified as people, and money constitutes speech (Citizens United v. FEC). At the same time, there is a right-ward and fascist political shift across the globe, with people like Espriella elected in Colombia, Trump in the United States, Modi in India, and a surge in far-right political parties in Europe, such as Reform in the United Kingdom, Danish People's Party and Denmark Democrats in Denmark, Alternative für Deutschland in Germany, and Front Nationale in France. This political shift is accompanied by an increasingly authoritarian shift in the corporate world: the eight most valuable companies on the NASDAQ account for roughly a third of the value of the stock exchange, and unequal wealth distribution between individuals is continually getting worse — the rich are getting richer and the poor are getting poorer. At the same time, advanced technologies for surveillance are being embedded widely into tools by the very companies that are engaging in efforts to bend regulation to their will. In this talk, we will discuss what the current trend of AI research is, what political developments accompany it, and where we are left as people. I argue that we as scientists are derelict in our duties to inform and educate the public — not out of negligence towards the public but a lack of awareness of the impact of the work that we have conducted. I call instead for slow and deliberate work towards social and technological futures that support democratic realities and sovereign people.

A pesquisa contemporânea sobre desinformação na área da Comunicação vem passando por uma mudança de perspectiva, deslocando o foco de peças isoladas de notícias falsas para a análise de ecossistemas complexos e infraestruturas digitais. Sob uma ótica cibernética, a desinformação é compreendida através de movimentos antiestruturais que exploram tanto as vulnerabilidades biológicas do cérebro quanto as "políticas do encanto", que mobilizam paixões e narrativas conspiratórias para criar realidades paralelas.

Vaccine disinformation spread widely on Twitter (currently X) in Brazil during the COVID-19 pandemic, shaping public debates and fueling distrust. To examine how such narratives organize online and induce hesitancy, we collected all 2,857,908 Portuguese-language tweets containing the word vacina (vaccine) between October 30th, 2020, and May 25th, 2021. We fine-tuned BERTimbau to detect disinformation, then used semantic similarity among tweets to identify a network of disinformation authors. Community detection revealed five well defined distinct clusters, each structured around a central narrative, with four containing a single prominent spreader. These findings show how disinformation circulates through identifiable communities that were fundamental for the spread of vaccine hesitancy in Brazil.