Collaboration Networks in Brazilian Computer Science

Subscribe

Apple

Google

Spotify

Overcast

Stitcher

RSS

-->

About<br>Home

During my bachelor's, I picked up Albert-László Barabási's Network Science book for a research project. The premise is elegant: protein interactions, power grids, social media feeds, financial markets, and many other complex systems can be modeled as graphs, with objects as nodes and relationships as edges. Once you have a graph, you can measure things: who is central, what clusters form, how fast information spreads, and whether the network has the "small-world" property where any two nodes are surprisingly close [1].

Scientific collaboration networks fell into this framework naturally and captivated me. I wanted to apply these ideas to Brazilian computer science to answer some questions that I had out of curiosity. The result is a peer-reviewed paper, Beyond Boundaries: Collaboration Networks and Research Output in Brazilian Computer Science, co-authored with André Vignatti, published in the XIV Brazilian Workshop on Social Network Analysis and Mining (BraSNAM) [2]. This post is my attempt to bring the findings out of LaTeX and into plain language, while adding some context about the research.

The Science of Measuring Science

Before the methodology, some framing. Bibliometrics is the quantitative study of scientific publications: how many papers are produced, who cites whom, how journals and conferences vary in prestige, and how those numbers change over time. A related but broader field, scientometrics , studies science itself as a social and epistemic system, asking how institutions, funding, geography, and policy shape what knowledge gets produced and by whom [3].

These fields have exploded in the last two decades, partly because open bibliometric databases made large-scale analysis feasible, and partly because funding agencies started using bibliometric indicators in evaluation processes. In Brazil, CAPES (which accredits graduate programs) and CNPq (the main research funding agency) both rely on metrics such as publication counts, citation rates, and collaboration breadth to assess researchers and programs.

Instead of counting papers and citations in isolation, Social Network Analysis (SNA) lets you ask structural questions: is this research community tightly clustered or spread out? Who are the "bridge" researchers connecting otherwise separate groups? How much does the network depend on a handful of key connectors? These questions cannot be answered by looking at individual papers; they require treating the entire community as a system.

Data Collection

There are a few options to collect publication metadata, some of them are open solutions like DBLP, Semantic Scholar, and OpenAlex, and also some commercial tools such as Scopus and Web of Science. All of these options enable us to access the publication metadata through API requests.

We decided to use OpenAlex for two main reasons. First, it has a detailed level of metadata, providing all the information we needed. For instance, it has institutional affiliations of the authors that support geographical analysis and systematic classification of publications within disciplinary subfields. Second, it offers great API documentation, especially when compared to other data sources, where the API data itself and the documentation are very confusing. Overall, OpenAlex is very easy to use, and the data we collected follows this structure:

We collected all data on March 31, 2025, and stored the raw dataset in CSV. The pipeline followed a classic ETL (Extract, Transform, Load) pattern:

Extract : Python scripts queried the OpenAlex API in batches of 25 records, organized by subfield and country to handle pagination constraints.

Transform : We cleaned duplicates (using DOIs and OpenAlex IDs as keys), removed entries without DOIs, and built co-authorship graphs using NetworkX. Edges were weighted by collaboration frequency and graphs were exported as GEXF files.

Load : Statistical visualizations were produced with Seaborn and network visualizations with Gephi.

All code and datasets are publicly available on our GitHub repository.

Global Landscape

Brazil ranks 12th globally in publication output with 76,184 publications and 447,919 citations. This positions Brazil in the middle tier of global research productivity, yet Brazil's citation ratio of 5.88 falls considerably short of high-impact nations like Australia, Great Britain, and the United States.

Rank<br>Country<br>Publications<br>Citations<br>Citations/Paper

🇨🇳 China<br>694,103<br>8,280,834<br>11.93

🇺🇸 United States<br>474,474<br>9,590,230<br>20.21

🇮🇳 India<br>311,644<br>2,224,750<br>7.14

🇮🇩 Indonesia<br>266,047<br>755,078<br>2.84

🇩🇪 Germany<br>141,044<br>1,960,782<br>13.90

🇬🇧 Great Britain<br>140,019<br>3,026,576<br>21.62

🇯🇵 Japan<br>103,265<br>799,048<br>7.74

🇫🇷 France<br>92,131<br>1,011,787<br>10.98

🇨🇦 Canada<br>84,076<br>1,562,176<br>18.58

10<br>🇷🇺 Russian Federation<br>83,214<br>363,210<br>4.36

11<br>🇮🇹...