A Telegram (multi-chat, multi-language corpus) dataset of 111,894,575 messages from 6,775 chats with 350,267 users and 99,459,189 reactions, collected by TeraGram Project Team via Telegram API. Each record contains message ID, chat ID, user ID, message type (PHOTO/VIDEO/TEXT/etc.), view count, forward count, reply and forward graph edges, user flags (bot/fake/scam/verified/premium), emoji reactions per message, and chat-level language detection. TeraGram is a large-scale Telegram dataset with 111,894,575 messages spanning 6,775 chats, including full reply and forwarding graph structure, 99,459,189 emoji reactions, and privacy-preserving encrypted user identifiers. Suitable for Telegram network analysis, information diffusion, bot and fake account detection, engagement modeling, multilingual content analysis, forwarding chain research.
TG-MOISES, a corpus of 6,805,925 messages from 490 public Telegram channels associated with conspiracy theories, anti-vaccine movements and far-right or nationalist narratives, in English, Spanish, Turkish, Lithuanian, French, German and other languages, collected by the MOISES project team through the Telegram API and released on Zenodo. 3,273,829 of the messages carry cleaned text; for the rest the export holds only an ID, the channel and flags. 20,941 messages (from 464 channels) were annotated by hand on a five-dimension malicious-content taxonomy (Role, Tactic, Feature, Target and Vulnerability), giving 401,150 label rows; the package also has the per-channel message counts, language distributions and topic-model output (404 channels, 21,614 topics with keywords, a theme text and an embedding vector each). The data have no author or time field for almost every message: only 24,752 messages have a date, between 2020-11-28 and 2023-12-19.
