Team Ai
Datasetpublic

hausmer/truexa-comments

Truha audience comments 45,918 real audience comments from the public satirical news channel «Труха⚡️Україна» (June 30 – July 22, 2026), collected from posts and cleaned (ads, links, mentions, ultra-short fragments, duplicates and phone numbers dropped — 50,000 raw → 45,918 clean). The audience writes in both Ukrainian and Russian (a mix, not a clean split — a share of comments code-switch between the two), so the corpus carries language: [ru, uk]. These are the "taste anchor"… See the full description on the dataset page: https://huggingface.co/datasets/hausmer/truexa-comments.

sourceHugging Faceotherupdated 25d agoView on Hugging Face
0likes65downloads
Dataset Card

Truha audience comments

45,918 real audience comments from the public satirical news channel «Труха⚡️Україна» (June 30 – July 22, 2026), collected from posts and cleaned (ads, links, mentions, ultra-short fragments, duplicates and phone numbers dropped — 50,000 raw → 45,918 clean).

The audience writes in both Ukrainian and Russian (a mix, not a clean split — a share of comments code-switch between the two), so the corpus carries language: [ru, uk].

These are the "taste anchor" of the satirical register: short, live, punchy or ironic one-liners — exactly the style the `hausmer/truha-gemma4` LoRA was tuned on.

Structure

ColumnTypeDescription
post_idintid of the post the comment was on
idintTelegram message id of the comment
datestringISO-8601 UTC timestamp
senderintnumeric Telegram user id (no names/handles stored)
textstringThe comment

Privacy

The comments were public in a public channel. Only a numeric (non- reverse-resolvable) sender id is kept — no usernames, no phone numbers, no links, no @mentions.

License

Collected 2026-07-22 via the public Telegram API. Comment authors retain copyright of their text; this archive is a research/imitation corpus and does not claim ownership.