ownly-network/china-news-feed-sample

China Deletion Watch: what vanished from China's hot lists

This is a free sample with capped volume: at most 100 flags per day, a rolling 30-day window, and 30 news rows per day. It is not the full feed.

Get the full volume, hourly refresh and full evidence trails: Ownly China News Feed with Deletion Watch on Apify -- see the Apify listing for current pricing.

Browse today's deletions for free: https://ownly-network.github.io/china-news-feed-samples/

Interactive demo: What vanished from China's hot lists (Hugging Face Space)

What's in it

Weibo 微博, Baidu 百度, Zhihu 知乎, Toutiao 今日头条 and Bilibili 哔哩哔哩 hot lists, checked every few minutes since 2026-09-30. An item that abruptly vanishes while still trending gets flagged likely censored; an item that comes back is flagged returned. Peak rank, bilingual headline, and timing come with every flag.

Config	File(s)	Rows	What it is
deletion_flags_recent (default)	data/deletion_flags_recent.parquet / .csv	up to 2889 (<=100/day, last 30 complete ET days)	the rolling capped sample
vanished_recent	data/vanished_recent.parquet / .csv	subset of the above	items that vanished while hot and never came back (drop_type == left_list, includes "Likely censored")
daily	data/daily/*.parquet	<=100/day, full available history	one capped file per calendar day; append-only once published
cross_source_sample	data/cross_source_sample/*.parquet	<=30/day	broader multi-source news sample, round-robin across source type
weekly file (download)	censored_2026-11-01.json	60	latest weekly snapshot (current)
weekly files (download)	feed_zh_2026-11-01.json / feed_en_2026-11-01.json	50 each	bilingual hot-list headlines

Fields (deletion_flags_recent, vanished_recent, daily)

Column	What it is
flag_id	stable id for this flag
platform	weibo / baidu / zhihu / toutiao / bilibili / douyin / kuaishou
hot_list_name	bilingual hot-list name
title_zh, title_en	headline (English is machine-translated; blank if untranslated)
url	public hot-list/search URL for the item
first_seen_et, last_seen_et	first/last time our poller saw the item on the list
first_missing_et	first successful snapshot where the item was gone
peak_rank, peak_heat	best rank / highest heat reached
minutes_on_list	minutes between first and last appearance
drop_type	left_list (abrupt exit, never returned), soft_drop (gradual decline, top 3 only, never returned), returned (came back later)
status_label, status_label_zh	"Likely censored" / "Left the list" / "Returned" (疑似被审查 / 已离开榜单 / 已恢复)
source_type	always "Commercial platform hot list" in these files
flag_date	ET calendar date of first_missing_et; the data/daily/ partition key

Method

Snapshots every 2-10 minutes per list, diffed for abrupt exits. "Likely censored" comes from the live detector's own verdict, recovered from every surviving copy of its output (the live file alone only keeps a few days of history). "Left the list" rows are our own best-effort reconstruction for everything that verdict doesn't cover. Full writeup, including exactly how drop_type/status_label are decided and the per-day cap: METHOD.md on this repo.

Limits

• A drop from a hot list is not proof of deletion. "Likely censored" is a judgement from rank history and re-checks, not a confirmation.
• status_label == "Left the list" rows are a weaker signal than "Likely censored": no URL recheck, no hard evidence, just an abrupt or notable (top-3) exit.
• Headlines are machine-translated; many title_en values are blank.
• Coverage gaps: snapshot outages exist, and douyin/kuaishou currently have no successful direct snapshots at all (only excluded mirror feeds).
• This is a capped, free sample -- at most 100 flags/day, a rolling 30-day window, 30 cross-source rows/day. The full volume is the paid Apify feed above.
• No personal data: no user handles, commenter names, or contact info.

License

CC BY 4.0 for Ownly Network's compilation, labels and summaries. Headlines and the linked pages remain with their publishers. Publisher: China News Feed.

• Interactive demo: https://huggingface.co/spaces/ownly-network/china-deletion-watch
• Free daily page: https://ownly-network.github.io/china-news-feed-samples/
• Full feed: https://apify.com/china.unofficial/china-news-feed

Load it

from datasets import load_dataset
recent = load_dataset("ownly-network/china-news-feed-sample", "deletion_flags_recent", split="train")
vanished = load_dataset("ownly-network/china-news-feed-sample", "vanished_recent", split="train")
daily = load_dataset("ownly-network/china-news-feed-sample", "daily", split="train")

Need every deletion as it happens, through an API, webhook or MCP? → Run the Apify actor