-- KamilaKoronska - 19 Sep 2023

Mapping Post-Truth Narraties:

Understanding Brazilian narratives about the war on Telegram through prompting

Team Members

Giulia Tucci, João Guilherme


Links
  • Link(s) to video / poster / slides

Contents


1. Introduction

This sub-report is part of a broader project aiming to map problematic narratives about the Russian-Ukrainian war since the beginning of the new phase of the Russo-Ukrainian war in February 2022. ‘Problematic narratives’ in this instance refer to the substance of persuasion campaigns, whether coordinated or organic and participatory.

We seek to identify these narratives to expose them as such and make them available to verification specialists as journalists, fact-checkers, disinformation researchers, human rights activists and educators. Typically these specialists work on single posts, producing fact-check’s or debunks of each. These single stories, however, may be part of larger narratives as described above.

In this report, we present both the step by step we created to make possible the analysis of Telegram data employing the GarganText tool and the path followed to answer the questions focused specifically in Brazil, made during the first week of the Summer School.

2. Research Questions

  • Which strains of false information about the war in Ukraine emerge in Brazilian Telegram channels? Can we distill the main narratives?

  • How does a leading AI platform make sense of the main keywords taken from problematic channels and construct their most significant narratives?

3. Methodology and Initial Datasets

The sequence of steps followed to develop the analysis are represented in Figure 1. First, relying on Telegram channels that spread problematic content about the Ukrainian war, we sourced additional channels by capturing their incoming and outgoing mentions and forwarding using TGSTAT ( tgstat.com). Using 4CAT to access the Telegram API, we downloaded the channels’ posts from February, 2022 to July, 2023.

The channel seeds and corresponding table were the following:

Seed channels

@guerranaucraniaultimasnoticias

@Grupos_Atualidades

@noticiasww

@ucraniavsrussia

@ucraniaerussia

@oinformantstarday

@falandodeguerra1

@canalartedaguerra

Channel name

Channel Username

URL

Country

ULTIMA HORA - SOS.

@UltimaHoraSOS

https://t.me/UltimaHoraSOS

BRA

SELVA & AÇO

@selvaeaco

https://t.me/selvaeaco

BRA

Randao

@randaoBR

https://t.me/randaoBR

BRA

QUESTIONE-SE

@questioneseoficial

https://t.me/questioneseoficial

BRA

Mundo PrintE

@Mundo_Printe

https://t.me/Mundo_Printe

BRA

FreedomNews

@freedomnewsforyou

https://t.me/freedomnewsforyou

BRA

Fora do Sistema

@EsquerdaFora

https://t.me/EsquerdaFora

BRA

CONECTIUM ... Inscreva-se

@conectium

https://t.me/conectium

BRA

Gafanhotos do Mercado - ONG - OB, Forex, B3, Crypto etc. TODO MATERIAL AQUI É GRATUITO E SEM FINS LUCRATIVOS

@canalgafanhotosdomercado

https://t.me/canalgafanhotosdomercado

BRA

Brasilposting

@brasilposting

https://t.me/brasilposting

BRA

FIM DOS TEMPOS PROFETIZADO

@batalhando_pela_fe_pr_alexandre

https://t.me/batalhando_pela_fe_pr_alexandre

BRA

Random shit 🔌🔌

@corvustt

https://t.me/corvustt

BRA

GUERRA🚀RUSSIA/UCRÂNIA🗽

@guerranaucraniaultimasnoticias

https://t.me/@guerranaucraniaultimasnoticias

BRA

GUERRA RUSSIA X UCRÂNIA - ATUALIDADES

@Grupos_Atualidades

https://t.me/@Grupos_Atualidades

BRA

Notícias da guerra e do mundo Rússia e Ucrânia

@noticiasww

https://t.me/@noticiasww

BRA

🚨NEWS🚨 UCRÂNIA X RÚSSIA ‼️

@ucraniavsrussia

https://t.me/@ucraniavsrussia

BRA

‼️ Ucrânia AGORA ‼️

@ucraniaerussia

https://t.me/@ucraniaerussia

BRA

O INFORMANTE

@oinformantstarday

https://t.me/@oinformantstarday

BRA

Canal Falando De Guerra

@falandodeguerra1

https://t.me/@falandodeguerra1

BRA

Canal ARTE DA GUERRA

@canalartedaguerra

https://t.me/@canalartedaguerra

BRA

[deactivated]

@canalselvabrasiloficial

t.me/canalselvabrasiloficial

BRA

Canal Visão Estratégica

@canalvisaoestrategica

t.me/canalvisaoestrategica

BRA

Lucas Leiroz

@lucasleiroz

t.me/lucasleiroz

BRA

Voz da Nova Resistência

@novaresistenciabrasil

t.me/novaresistenciabrasil

BRA

Politicamente Incorreto I Love to Hate U 🔥🎡

@polincbr

t.me/polincbr

BRA

Semeando a discórdia

@SemeandoDiscordia

t.me/SemeandoDiscordia

BRA

We developed an R script to compile the .csv files, rename its columns, create additional information, filter by keywords, clean the text to create a dataset, making it compatible with GarganText input by using tidyverse, readr and stringi packages. This code was also used by the other groups in the project that analyzed other countries' Telegramsphere. We subsequently loaded those posts in Gargantext, the text mining and analysis software, in order to group the terms and find narratives.

The code was based on the following sequence of steps:

#Team's country

Register the files’ country to help us to organize the code outputs before inserting them into GarganText.

#Set the working directory to your csv files source

Considering that some countries have more than one csv file to convert, the code input is the address of the folder with the csv files to compile before processing.

#Finding all csv files in the folder and using a function into all files found

An identifier registers all the csv files in the working directory and inserts it in an original function made to compile all csv files in an object called Data.

#Creating a function to adapt the column names to fit into the GarganText template.

We wrote a code able to:

  • Rename the columns of the csv outputs from 4CAT to fit the GarganText template, column names and order.

  • Creates specific and separated day (‘Publication Day’), month (‘Publication Month’) and year (‘Publication Year’) columns to fit GarganText.

  • Filter keywords related to the themes of interest listed by each country, considering a wordlist.

  • Clean the text to remove special characters that could interfere in GarganText working.

  • Remove columns other than Publication Day, Publication Month, Publication Year, Authors, Title, Source and Abstract.

  • Define the threshold of lines of the output considering GarganText limits.

  • Write a csv containing the country name and the date of the analysis.

#Produce a GarganText output

Finally, the output csv is compatible with GarganText to all countries.

The code was used by the Brazilian, Polish, Chinese, Italian and Indian teams, with minor adaptations made by the Brazilian team for each case.

Using Gargan Text and the Brazilian corpus, we composed a network of words and chose the ones with higher network degree to make prompting queries to ChatGPT to understand how it makes sense of these sets of keywords. After giving the context, we politely prompted:

"We are doing research in which we collected messages from Telegram channels that disseminate information and disinformation about the war in Ukraine, built a network with the most used terms and ran a modularity algorithm to calculate which cluster each node (term) belongs to. I will write the words of each cluster here and ask you to name the disinformation narratives that these terms represent. Is it okay?"

After promising to do its best, ChatGPT named the terms for each cluster.

4. Findings