In this work, we present the “Telegram Monitor”, a web-based system that monitors the political debate in this environment and enables the analysis of the most shared content in multiple channels and public groups. Our system aims to allow journalists, researchers, and fact-checking agencies to identify trending conspiracy theories, misinformation campaigns, or simply to monitor the political debate in this space along the 2022 Brazilian elections. We hope our system can assist the combat of misinformation spreading through Telegram in Brazil. The following link contains a brief description about the aforementioned system: https://bit.ly/3l4xNrF

CCS Concepts: CCS Concepts: • Human-centered computing → Empirical studies in collaborative and social computing ;

Keywords: Keywords: Telegram , instant messaging platforms , misinformation , fake news , politics ACM Reference Format: Manoel Júnior, Philipe Melo, Daniel Kansaon, Vitor Mafra, Kaio Sa, and Fabricio Benevenuto. 2022. Telegram Monitor: Monitoring Brazilian Political Groups and Channels on Telegram. In Proceedings of the 33rd ACM Conference on Hypertext and Social Media (HT '22), June 28-July 1, 2022, Barcelona, Spain. ACM, New York, NY, USA 4 Pages. https://doi.org/10.1145/3511095.3536375

ACM Reference Format: ACM Reference Format: Manoel Júnior, Philipe Melo, Daniel Kansaon, Vitor Mafra, Kaio Sa, and Fabricio Benevenuto. 2022. Telegram Monitor: Monitoring Brazilian Political Groups and Channels on Telegram. In Proceedings of the 33rd ACM Conference on Hypertext and Social Media (HT '22), June 28-July 1, 2022, Barcelona, Spain. ACM, New York, NY, USA 4 Pages. https://doi.org/10.1145/3511095.3536375

Instant messaging platforms have become one of the main ways to communicate with close friends and also find communities of people who share the same interest, creating large public groups and channels, gathering thousands of users around a topic of discussion. Telegram ranks among one of the most popular with 550 million monthly active users around the world [21] and has seen a huge growth in its user base. Together with their biggest competitor, WhatsApp, these platforms operate a network that exchanges billions of messages every day [25]. Unfortunately, instant messaging apps also became places for the spreading of misinformation campaigns [19, 20], conspiracy theories [9], terrorist content [22], and hate speech [2]. These problems, along with public pressure on companies such as Facebook, motivated the establishment of measures to make it difficult to spread content quickly, such as limiting the forwarding of messages [15] and banning accounts that violate the platform's terms of use. In this scenario, other messaging platforms, including Telegram itself, end up emerging as an unmoderated alternative in which users are free to discuss whatever they want without any restrictions.

Although Telegram gained some prominence and lot of public groups are being created everyday on the platform [10], little has been investigated about this service. Unlike other social media, it is a hard task to have any information about the huge content circulating on this platform. Due to the closed nature of these systems, not even the most discussed trends or topics are known to those who intend to analyze this environment.

This work is built on the creation of a Telegram public group monitoring system – the Telegram Monitor. With this system, we seek to help better understand the ecosystem of messaging applications in Brazil, creating an interface that brings out some of the most shared media and messages in political groups and public channels on the platform, facilitating the investigation of the content that circulates on this network. Telegram has already attracted some attention in Brazil for the volume of misinformation found on the platform [11]. The Telegram Monitor is an online system that displays, in a period of time selected by the user, the most shared media on these days in the investigated public groups, similar to other initiatives with the same proposal, but for WhatsApp [14]. This system is updated on a daily basis with images, audios, videos, and messages shared, allowing users to discover the most popular topics that are present on the platform at a given a period of time. This can help many researchers and journalists that have access to the system to observe how these platforms are being used and easily identify false stories getting viral on its network, working as a tool to fight against misinformation. The system's architecture integrates an extensive collection of content from public political groups in Brazil, with the storage and processing of all media shared in these communities. To understand the popularity of each content, the collected media are grouped by similarity, based on a perceptual hash algorithm that generates a unique identifier for each content, thus being able to aggregate similar content and quantify how many times each one was shared by the users. Finally, this content is processed on a daily basis, ranking each type of media by popularity and displayed in an online access system through a URL with a link to the system: http://www.monitor-de-telegram.dcc.ufmg.br/. Through an account with a username and password, it is then possible to access all the data and browse the days, exploring what was shared in the monitored groups and channels.

Telegram is a cloud-based, cross-platform instant messaging application, where the accounts are validated by a phone number. With this service, users can send messages to their contacts and participate in chat groups of up to 200,000 members and unlimited size channels. They can also invite others to their groups/channels by sharing an invitation URL link. Our methodology uses data collected from these public groups and channels. The system monitors groups and channels for political discussion in Brazil. Those are operated either by individuals with an interest in the topic, and can be freely accessed by anyone with the invite link they share. They can also be found in the search feature built into the Telegram apps. In this section, we describe all methodology behind the development of Telegram Monitor. The main architecture is explained through the flowchart in Figure 1. In this Figure, we have an overview of all the steps performed to build the online system: (i) A phone with a valid Telegram account is set. (ii) We find a set of relevant groups and channels on the Web to monitor. (iii) A script extracts and structures all messages received. (iv) Media files (image, video and audio) are downloaded. (v) Another step is responsible for parsing these files and grouping the messages by similarity. (vi) Merged messages are saved containing the total number of times they were shared. (vii) The online system accesses the database and displays it in the interface according to the user's filters. Next, we go through each step separately giving details on how they were deployed.

Data Collection: The first step in gathering data on Telegram is to define which are the groups and channels of interest to the collection. For this work, we focus on monitoring Brazilian political public groups and channels. A common practice for gathering groups in messaging apps is to use keywords along with the group invitation link pattern to identify potential groups and channels of interest for collection [4, 7, 12, 14, 20]. Telegram group invite links follow the pattern https://t.me/joinchat/<GroupID> or https://telegram.me/<GroupID>. In this way, we manually created a list of terms related to the Brazilian political context. Then, we searched the keywords of the list together with the invitation URL pattern on search engines (Google) and social media (Twitter and Facebook) to find posts containing public groups of our topic. In addition to this, Telegram has a group search system within the application itself, where terms were also searched to find more groups.

With this process, we identified and joined a total of 232 public groups and channels related to politics with thousands of members. Some of the selected channels have more than a million members. It is interesting to note that around 66% of the groups and channels found have more than 256 members or subscribers (maximum size of a WhatsApp group). This gives us a hint of how big is the audience that information spreading can reach on those Telegram groups/channels which is much bigger than WhatsApp groups maximum capacity.

Content Processing: The Telegram platform has an official API^1^ in which it is possible to obtain conversation data, in addition to performing actions on behalf of the user, such as joining groups and sending messages. We used this API to gather historic data from all the groups we were able to join. This data includes text messages, images, videos, audios, documents and any other type of content that the platform supports in its official clients. Finally, to create a system that allows one to assess the most popular content shared in a period, we separate the data by day and, for each day, we rank each type of media according to its popularity on the selected date.

To be able to track and count how many times the same piece of content has been shared (measuring its popularity), we need to do the whole process of tracking, grouping, and counting the data, by finding all copies of that message (image, audio or video) among others. To find copies of the same content, we use the pHash – perceptual hashing algorithm [16] on images. This process generates a visual hash that works as a kind of “fingerprint” of the content, making it possible to compare two images. For audio and video we also calculate a hash from the checksum (MD5) that generate a unique value for each different file. Similarly, we also use the Jaccard index to compare text messages and group them. With the hashes, we can detect content that is identical or very similar and group them. As a result, it is possible to calculate aggregated information about each piece of content shared, such as determining its popularity, counting in how many groups it appeared and how many different users sent it as shown in Figure 2.

Ethics: Telegram Monitor gathers a considerable amount of data from many Telegram groups and users. To ensure users’ privacy, we do not share or disclose any personally identifiable information, such as phone numbers or usernames. To prevent misuse, even of aggregated information, we also limit access to our system to a restricted number of people, through an account with a password. In addition, they are also informed about the limitations of the data and the potential bias present in our system. As we only use publicly available Telegram groups and official tools provided by the system's API, our data collection does not violate Telegram's terms of service.

Online Interface: The Telegram Monitor has restricted access through an account with a username and password. Once logged in, the user can select a day (or a period of time with several days), as shown in Figure 3, and the system displays a ranking of the most popular content of the selected period. After picking a date, it is possible to find the most shared content according to the type of media (images, videos, audios, or text). When choosing the media, the content will be ranked based on how many shares it had in that period, sorted in descending order by the number of shares. An example of the system interface can be seen in Figure 4. For each content displayed, it is also possible to see more details of it, with information about how much it has spread. In detail, we have the total shares, the number of groups in which this content appeared and the number of different users who sent it. It is also displayed a list of names of the groups in which it has been shared. Finally, there is a button that searches for the image on Google to check if that same image appears elsewhere on the Internet.

Our monitoring system has already attracted some attention as a tool to combat misinformation. Particularly, a key challenge for fact-checking organizations and investigative journalism in this space is to avoid giving too much attention to conspiracy theories and fake stories that are not getting viral. The concern here is not only to work on something irrelevant, but also to boost the attention to some unpopular misbeliefs. The content displayed at our monitoring system allows fact-checkers to quickly get a sense if a claim is getting viral at least in this specific system. In this sense, our university has already become a technical partner of Comprova [18], a project for collaborative fact-checking in Brazil, recognized by the International Fact-checking network, with more than forty Brazilian media portals. Other key fact-checking agencies in Brazil including Lupa [1], Aos Fatos [3], and Estão Verifica [6] already access our system. Moreover, we gave access to our system to dozens of journalists with an editorial line and some of them already used our system, explicitly mentioning it as a data source for their investigation [8, 13, 23, 24].

Finally, it is important to mention that Telegram has ignored Brazilian authorities in different situations [17], which raises concerns about the abuse of Telegram by misinformation campaigns during the upcoming elections. The Superior Electoral Court (TSE), which is responsible for conducting the Brazilian 2022 elections, has also partnered with our project [5]. Our system may represent a way for TSE to get known about possible misinformation campaigns attacking the democrat process through Telegram and take proper actions if necessary. As future work, we aim at improving the system's capabilities to provide more valuable information about the content displayed as well as more detailed information on aggregated data via reports to journalists and fact-check agencies.

This work was partially supported by research grants from CNPq, FAPEMIG, and FAPESP.