2023 Shortlist
As eleições brasileiras em 11136 vídeos personalizados | The Brazilian elections in 11136 customized videos
Entry type: Single project
Country/area: Brazil
Publishing organisation: g1/globo
Organisation size: Big
Publication date: 2022-10-02
Language: Português
Authors: Coordination: Felipe Grandin, Guilherme Gomes, Mariana Mendicelli, Renato Franzini, Ricardo Gallo, Tatiana Caldas e Thiago Madeira
Design: Guilherme Gomes, Fernando Lima, Vitória Coelho, Verônica Medeiros e Max Francioli
Big Data & AI: Thiago Madeira, Igor Costa, Rafael Muniz, Jozimar Back, Christian Polo e Mari Gibara
Production, review and auditing: Ardilhes Moreira, Laura Naime, Braulio Lorentz, Roberto Peixoto, Raphael Martins, Marília Neves, Caroline Prado, Marta Cavallini, Luisa Belchior, Bruna Miato, Victor Cinzento, Emily Cruz, Raphael Martins, Arthur Stabile, Fernanda Calgaro, Gustavo Morais, Matheus Moreira, Cauê Muraro, Felipe Gutierrez, Carolina Lorencetti, Julia Putini, Aline Macedo, Gabriela Sarmento, Cesar Soto, Matheus Rodrigues, Gabriel Croquer, Darlan Helder, Anaísa Catucci, Geisy Negreiros, Tácita Muniz, Cau Rodrigues, Heliana Gonçalves, Roberta Batista, Vivi Leão, Fabiana Figueiredo, Núbia Pacheco, Valma Silva, Itana Alencar, Maiana Belo e Natally Acioli, Sidney Magno Novo, Fabiana Oliveira, Viviane Lopes, Rubens Guilherme Santos, Fernanda Lourenço Carloni, Gladys Peixoto, Cíntia Araújo, Rafaelle Fróes, Geisa Almeida, Liliane Cutrim e Marcia Carlille, Anderson Viegas, José Câmara, Rafaela Moreira, Renata Barros Debora Ricalde, Thais Libni, Pedro Mathias, Caroline Mesquita, Mariana Mouro, Arthur Sobral, Taymã Carneiro, Taiguara Rangel, Krys Carneiro, Diogo Almeida, Bruna Couto, Iara Alves, Jhonathan Oliveira, Mendonça, Bruno Marinho e Katherine Coutinho, Caroline Oliveira, Maria Romero, Catarina Costa, Andrê Nascimento, Laura Moura, Josiel Martins, Bibiana Dionísio, Letícia Paris, Caio Budel, Ana Kruger, Mariah Colombo, Barbara Hammes, Gilvana Giombeli, Rafael Machado, Gabriel Bukalowiski, Breno Antunes, Alanis Barbosa, Caroline Maltaca, Lucas Ravel, Fillipe Borba, Fernanda Zauli, Augusto César Gomes, Felipe Galdino, Igor Jácome, Leonardo Erys, Valéria Oliveira e Samantha Rufino, Lilian Lima, Gabriela Clemente, Andressa Santos, Juliana Borgmann, Juliana Lisboa, Gustavo Foster e João Pedro Lamas, John Pacheco, Clarissa Batistella, Caroline Borges, Sofia Mayer e Joana Caldas, Joelma Gonçalves, Isabelle Marques, Demetrius Oliveira, Gustavo Rodrigues, Leonardo Barreto, Vilma Nascimento, Patrício Reis e Patrícia Lauris, Fabiano Rodrigues, Felipe Santos, Filipe Ferreira, Thaís Contarin, Lucas Soares, Franco Jr, Cristiane Rodrigues, Marina Pereira, Michelly Oda, Fran Ribeiro, Fabiano Rodrigues, Felipe Santos, Filipe Ferreira, Thaís Contarin, Cristiane Rodrigues, Marina Pereira, Michelly Oda, Fran Ribeiro, Fabiano Rodrigues, Felipe Santos, Filipe Ferreira, Thaís Contarin, Hayale Guimarães, Vinicius Lima, Ana Marin, Fabio Rodrigues.

Biography:
The project was a team effort involving journalists, designers, developers, video editors and many profissionals from g1/globo around Brazil.
Project description:
g1 generated and published 11136 customized texts and videos with Brazilian two rounds election results hours after the votings in October 2022. Each had the results for a specific city, including votes and percentages for the President, Governor, and other elected officials. Delivering fast and detailed information for voters was crucial to counter coordinated attacks on the voting systems and false claims of election fraud that had been spreading before the ballot. It was the first time a news outlet released this number of automated election videos in such a short time. The content reached millions of people.
Impact reached:
The fast release of 5568 automated texts and videos with results of each Brazilian election round in October 2022 played a significant role in g1/globo effort to inform voters during the fiercest presidential election in the country’s history. At that time the fear of massive countrywide protests to overturn the elections was real, and the quick and efficient way the press and Electoral Justice inform the public about the results are tought to have played a key role in preventing demonstrations to scale.
g1/globo managed to deliver text and video with specific data for each Brazilian city, making the results more accessible and easy to search. One of the goals in automating content was to reach very small towns that compose two-thirds of Brazilian cities and are not used to get customized content from the press.
The strategy of releasing automated content with citywide results was proven right when false messages started to spread showing fake presidential vote counts for Bolsonaro or Lula in specific cities. They could be easily disproven by the automated texts and videos.
Millions of people viewed the automated texts and videos with Brazilian election results.
Techniques/technologies used:
g1/globo already had experience publishing automated texts in 2020 municipal elections – with great results. Automated videos were the next step to ensure a broader reach, specially on social media. But the format presented many technical challenges, such as processing power, data quality and CMS integration.
To tackle those technological challenges, a complex data processing pipeline was developed, relying most on cloud technologies using Google Cloud Platform (GCP). The advantage of this approach is the possibility of scale by demand besides the resilience of the operation. All data was landed in the JSON format, and the buckets service from Google Cloud Storage was used to store them, the JSON files were converted to AVRO format which is a better format suited for querying and processing with Python scripts, and beyond that, the data is stored in Big Query tables to be consumed by front-end applications such as dashboards, the pipeline flow management is handled by the Dataflow.
For the video rendering process, virtual machines from Google Compute Engine were used to build a render farm, running on Windows OS. The rendering processing relies upon Adobe After Effects. It is automated by the Nexrender framework, which is written in Javascript and is responsible for “gluing” all the data to the videos. At the end of the rendering process, the videos are encoded through the FFMPEG framework.
For the data quality insurance, an OCR layer is added to the pipeline flow, which is responsible for extracting information from each frame of the video and guaranteeing that it matches the correct data of each city and candidate polling numbers. This system is developed using the Pytesseract framework for the OCR part, and the OCV library for image treatment, such as grayscale tunning and noise reduction.
Context about the project:
Brazil held in October 2022 the fiercest presidential election in its history. In a runoff Brazilians ousted President Jair Bolsonaro and elected former President Luiz Inácio Lula da Silva after months of disinformation campaigns, coup threats, and violence among voters. Coordinated attacks on the voting systems and false claims of election fraud had been spread by Bolsonaro supporters years before the ballot.
There was a widespread notion among electoral authorities and the press that after the votes were counted time was of the essence to disperse any attempts to overturn the elections. Delivering the election results in a fast, detailed, and reliable way was essential to assure the public confidence in the voting systems and Democracy itself.
g1/Globo has been publishing for years detailed election data supplied by the Electoral Justice in charts and interactive pages as soon as they are available, with great success. But not all people find charts friendly and accessible. Many prefer to read a text or watch a video. Besides, specific results for each city are not easily found in search engines.
With that in mind, g1 started publishing automated texts with election results for each of the 5568 Brazilian cities in the 2020 municipal election minutes after the numbers were released. The response was very positive and the content reached a broader public.
The logical step was to adopt one more time text automation to amplify content production in the 2022 elections. Given the political climate though, reaching the maximum possible number of people was not a goal but the top priority. The text was not enough. It had to have a video.
Nonetheless, automating video has a series of challenges, such as processing power, data quality, and CMS integration. And it had never been done at this scale by a news outlet, at least in Brazil.
The project took nine months until it was ready to go live on election day.
It involved different departments and suppliers, a series of test to ensure the data was going to be correct and passed by sampling audit before publishing and a complete review after that found no data errors.
The content was viewed by millions of Brazilians after the election and helped inform the public with speed, precision and customization.
What can other journalists learn from this project?
The use of tools to automate the creation of news texts and videos can massively improve speed and volume of a newsroom production while also allowing news organizations to offer customized content that otherwise would be impossible to make. That can prove to be specially relevant when there is a need to adress many small groups of people with specific information, such as desagregating national election results to a county level.
Project links: