First propaganda detection mechanism for Telegram

Table of Contents

The new method developed by EPFL researchers in collaboration with colleagues in Germany, is efficient, fast and inexpensive. This means that it could help moderators in the fight against disinformation on Telegram

Image by motionstock from Pixabay

In an article signed by Tanya Petersen we read that **disinformation campaigns and propaganda take place not only in traditional social media, but also in messenger services such as WhatsApp, Telegram and Signal. And yet, there’s been little scientific research into the latter to date.

Now, researchers affiliated with both EPFL’s Security and Privacy Engineering Laboratory (SPRING) in the School of Computer and Communications Sciences and the Max Planck Institute for Security and Privacy with colleagues from Ruhr University in Bochum, Germany, have developed a new method to track down accounts that systematically spread propaganda in Telegram channels.

In some countries, such as Russia and Ukraine -EPFL said-, Telegram is one of the main information platforms. Content moderation on the platform is in the hands of the people who run a group or channel. There’s no independent fact-checking authority. Moderators are free to set their own rules in their groups and channels, and they often resort to removing unwanted content manually. Some users also use software for this purpose, but such tools are typically based on very simple criteria, such as detecting obscene language.

Russian and Ukrainian propaganda networks identified
#

As outlined in their paper, the researchers analyzed 13.7 million comments from 13 Telegram channels with a focus on politics or news. 1.8 percent of them turned out to be propaganda. The majority was spread by a pro-Russian network, which had posted up to 5 percent of all messages in some channels. The researchers also uncovered a smaller pro-Ukrainian propaganda network.

How propaganda accounts operate on Telegram
#

Their detection mechanism is based on the responses that the propaganda accounts give to the accounts of ordinary users. As the researchers showed, propaganda accounts on Telegram don’t initiate conversations, but respond to comments from other users that contain certain keywords, such as Putin or Zelensky.

“One of the most notable characteristics of propaganda accounts is that they post messages with the same wording in different places, sometimes across different channels. While regular accounts post unique messages, propaganda accounts form large networks that repeatedly disseminate the same content,” explained Associate Professor Carmela Troncoso, Head of the SPRING Lab.

The researchers used this feature to program an automated detection mechanism. It was able to recognize propaganda based on a single comment with a hit rate of 97.6 percent, which is 11.6 percent better than a human.

In this diagram, each dot represents a Telegram account. A line indicates that an account has replicated a comment from another account with the same wording. The diagram shows that propaganda accounts (right) behave very differently to normal accounts (left). They create networks that repost the same content over and over again. © 2025 EPFL / Ruhr University

Human moderation has varying degrees of success
#

As part of their study, the group also analyzed how efficiently moderators combat propaganda. The results varied greatly from channel to channel. While some news channels managed to identify and delete almost 95 percent of propaganda, the rate on other channels was just under 20 percent. According to the channel operators, removing propaganda requires a great deal of manual effort. The new detection tool could reduce both time and effort significantly.

“Content moderators are not available around the clock and being constantly exposed to propaganda can also affect people’s mental health. This is why we hope that our algorithm will help with moderation on Telegram going forward,” concluded Klim Kireev, a doctoral student in the SPRING Lab and first author of the study.

Citation
#

The paper “Characterizing and Detecting Propaganda-Spreading Accounts on Telegram” was presented last week by Professor Rebekah Overdorf from Ruhr University Bochum, previously at EPFL, at the 34th USENIX Security Symposium, the authors were awarded with a Distinguished Paper Award for their work. Authors: Klim Kireev, Yevhen Mykhno, Carmela Troncoso & Rebekah Overdorf.
NdelT.: Links to Wikipedia were done by Notaspampeanas.