Eleven Years of BRACIS: A Meta-Scientific Study of the Brazilian Conference on Intelligent Systems
This meta-scientific study analyzes eleven years (2015–2025) of the Brazilian Conference on Intelligent Systems (BRACIS) to characterize its evolving research topics, author demographics, citation patterns, and growing openness practices, while highlighting the significant accessibility barriers caused by paywalled proceedings.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine a giant, bustling library where scientists from all over a country gather to share their latest inventions and discoveries. This isn't just any library; it's the home of the Brazilian Conference on Intelligent Systems, or BRACIS. Think of it as the national stage for Artificial Intelligence (AI) in Brazil, a place where researchers present their work on teaching computers to see, think, and solve problems. To understand this paper, you need to know three simple things. First, bibliometrics is just the art of counting and measuring things in science, like tallying how many books a writer has published or how often other people read them. Second, citations are like high-fives in the research world; when one scientist mentions another's work in their own paper, it's a citation, showing that the original idea was helpful. Finally, openness is the practice of sharing the "recipe" behind a discovery—like posting the code or data online so anyone can try it themselves. We care about these things because they tell us who is doing the work, what they are working on, and whether their ideas are actually making a difference or just sitting on a shelf.
This paper is a ten-year detective story about the BRACIS conference, looking at 1,066 papers published between 2015 and 2025. The authors acted like digital archaeologists, digging through the conference records to answer three big questions: What topics are they studying? Who is showing up to the party? And which papers are getting the most attention?
They found that the party has changed a lot over the decade. In the early years, the research was mostly about building new algorithms from scratch, like inventing a new type of engine. But recently, the focus has shifted toward building "models" (like pre-trained AI brains) and testing them in real-world situations. A huge new trend has exploded since 2020: research on Large Language Models (LLMs), the kind of AI that writes text and chats with us. Before 2020, there were zero papers on this; by 2024, nearly one in five papers (19%) was about LLMs. The conference also serves as a special home for research written in Portuguese, a topic that bigger international conferences often ignore.
When it comes to the people, the community looks a bit like an hourglass. Most authors (80.5%) show up just once, like a one-time guest at a festival, while the institutions (universities and labs) are the steady hosts that keep coming back year after year. The most active region has been the Northeast, which grew faster than any other part of the country. The top university is USP, which published 190 papers, but the industry players are mostly Brazilian banks and tech startups, with Itaú Unibanco leading the pack.
The most surprising part of the story is about popularity. The authors discovered that citations are incredibly concentrated, like a rock concert where 1% of the songs get 27% of the applause. A tiny handful of papers, mostly about Portuguese language AI tools like "BERTimbau" and "Sabía," carry almost all the weight. In fact, if you remove those few superstars, the "best paper" nominees don't actually get more citations than the average paper.
The paper also investigated whether being "open" helps a paper get noticed. They checked if sharing code (artifacts) or posting a draft on a public server (arXiv) made a difference. They found that sharing code didn't seem to boost citations, but having an arXiv preprint did. However, the authors are careful to say this is just a connection they found, not proof that posting online causes more citations. The biggest hurdle they identified is that the conference papers are locked behind a paywall, like a library with a high fence. Only 7.4% of the papers have a free preprint version, meaning most of this brilliant work is hard to find unless you belong to a university that pays for access. The authors suggest that if the conference moved to an open-access model, more people could finally read and build upon these ideas.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.