Datasetpublic

hausmer/dvach_chat

Двач чат — group chat (Telegram) The group chat that runs alongside the public «Двач» channel (hausmer/dvach). While the channel is a single-author broadcast, this is a multi-participant group: every message carries the name of the person who posted it. This corpus is a text-only snapshot of the whole group history as exported by a member, with media stripped. This corpus contains 568516 text messages from 2022-09-02T15:35:10+02:00 to 2026-09-15T21:18:52+02:00, posted by 4851… See the full description on the dataset page: https://huggingface.co/datasets/hausmer/dvach_chat.

sourceHugging Faceotherupdated 10d agoView on Hugging Face

0likes63downloads

Dataset Card

Двач чат — group chat (Telegram)

The group chat that runs alongside the public «Двач» channel (hausmer/dvach). While the channel is a single-author broadcast, this is a multi-participant group: every message carries the name of the person who posted it. This corpus is a text-only snapshot of the whole group history as exported by a member, with media stripped.

This corpus contains 568516 text messages from 2022-09-02T15:35:10+02:00 to 2026-09-15T21:18:52+02:00, posted by 4851 distinct participants (547410 of them are replies to another message). Rows are split into 6 parts of up to 100,000 messages each so the viewer can page through the full history.

Columns

column	type	description
`id`	int64	Telegram message id (unique across the whole corpus)
`date`	string	message timestamp, ISO, local tz as exported (+02:00)
`author`	string	the participant's name, verbatim as displayed in Telegram
`reply_to`	int64?	id of the message this one replies to (`null` otherwise)
`chat`	string	always `"dvach_chat"`
`text`	string	full message text (tags stripped, `<br>` → newline)

What this is (and isn't)

This is a private group chat, not a broadcast channel. Each row is one message from one participant. The author field is the display name shown in Telegram, reproduced verbatim (the group has ~8k participants, most using pseudonyms; a subset use real first+last names). Messages are in the 4chan-aggregator / trolling tone of the sibling channel: vulgar, absurd, offensive. Reader discretion strongly advised; do not expose to general-audience settings.

Caveat

Content is © its respective authors. This is a research snapshot exported by a group member, published as-is; it is not an official or sanctioned release. Media (photos, files) was stripped, so formatting and attachments are lost. Deleted participants appear under Deleted Account.

Usage

python

from datasets import load_dataset
ds = load_dataset("hausmer/dvach_chat")
print(ds["part-000"][0])