What repeated sampling of hot lists shows, 4 to 31 Oct 2026

The question this answers for a buyer

What does a daily record of platform hot lists tell me that looking at today's ranking cannot?

The exact question and range

Across the Baidu, Weibo and Tencent WeChat hot lists between 2026-10-04 and 2026-10-31, how long do topics stay on a list, how many appear in only one collection window, and how many appear on more than one platform?

Date range

2026-10-04 to 2026-10-31, inclusive

Frozen

2026-11-02 01:46 UTC

Records

9,618

Result

15,087 observations pulled collapse to 9,618 platform-topic pairs. 67.7% appeared in only one collection window, so a single current ranking shows the topic but not whether it will last, or whether it already had. 26 normalized titles appeared on all three platforms.

• Topics seen in only one collection window (all platforms)67.7% (6510 of 9618)A collection window is a run of observations with no gap over 30 minutes.
• Baidu: topics seen in one collection window61.8% (2253 of 3647)107 collection windows in the pulled data.
• Baidu: median lifespan0.0 hoursFirst to last observation, counting one-window topics as 0. Among topics seen in 2+ windows: 4.5 hours.
• Weibo: topics seen in one collection window88.2% (4124 of 4678)107 collection windows in the pulled data.
• Weibo: median lifespan0.0 hoursFirst to last observation, counting one-window topics as 0. Among topics seen in 2+ windows: 4.5 hours.
• Tencent WeChat: topics seen in one collection window10.3% (133 of 1293)107 collection windows in the pulled data.
• Tencent WeChat: median lifespan11.5 hoursFirst to last observation, counting one-window topics as 0. Among topics seen in 2+ windows: 11.5 hours.
• Longest persistence: 这些关于中风的误区,不要再信了!10 windows over 2 daysTencent WeChat; best rank 11; 25.3 hours between first and last seen.
• Longest persistence: 我国将加强困难老人探访关爱10 windows over 2 daysTencent WeChat; best rank 8; 22.5 hours between first and last seen.
• Longest persistence: 致敬冼星海:把家国写进音符10 windows over 2 daysTencent WeChat; best rank 2; 22.5 hours between first and last seen.
• Longest persistence: 两个人的婚礼变成一群人的狂欢9 windows over 2 daysTencent WeChat; best rank 2; 21.8 hours between first and last seen.
• Longest persistence: 官方介绍五中全会最重要成果9 windows over 2 daysTencent WeChat; best rank 1; 21.8 hours between first and last seen.
• Exact-title overlap: Baidu and Weibo162Topics with an identical normalized title on both. Exact matching undercounts: one event is usually worded differently on each platform.
• Exact-title overlap: Baidu and Tencent WeChat629Topics with an identical normalized title on both. Exact matching undercounts: one event is usually worded differently on each platform.
• Exact-title overlap: Weibo and Tencent WeChat46Topics with an identical normalized title on both. Exact matching undercounts: one event is usually worded differently on each platform.
• Exact-title overlap: all three platforms26Identical normalized title on every platform.

Reproduce the documented result

One anonymous request with a fixed date range and at most 100 rows. Running it returns the rows counted below; the digest is the SHA-256 of their sorted ids joined by a newline.

curl -sS 'https://china-snapshot.com/api/datasets?dataset=social&mode=topics&source=Weibo%20Hot%20Search&from=2026-10-31&to=2026-10-31&limit=100'

Request

Weibo topics on 2026-10-31 (mode=topics)

Rows returned

100

ids_sha256

930f957acab0eb55e7e2da3b7c5c795a318f3ac71e17426d5656d24b226b32f6

Further pages

Yes. The documented result is this first page only.

import hashlib
import requests

items = requests.get("https://china-snapshot.com/api/datasets?dataset=social&mode=topics&source=Weibo%20Hot%20Search&from=2026-10-31&to=2026-10-31&limit=100", timeout=60).json()["items"]
digest = hashlib.sha256("\n".join(sorted(i["id"] for i in items)).encode()).hexdigest()
print(len(items), digest)  # documented: 100 930f957acab0eb55e7e2da3b7c5c795a318f3ac71e17426d5656d24b226b32f6

Sources

• China Snapshot public datasets API (social) — Repeated observations of publisher hot lists.
• Baidu: Baidu Top Realtime — 5457 observations pulled
• Weibo: Weibo Hot Search — 5350 observations pulled
• Tencent WeChat: Tencent WeChat Hot Topics — 4280 observations pulled

Known gaps

• The reproducible request returns the first 100 rows of a longer result (the API answered with next_cursor). result.row_count and result.ids_sha256 describe exactly that response, not the longer result.
• A topic is identified by its normalized title (leading rank and extra whitespace removed, Unicode NFKC). A story reworded between collections counts as a new topic, so persistence is understated; exact-title overlap across platforms is likewise a floor.
• Collection windows are inferred from observation timestamps (a gap over 30 minutes starts a new window). Missed or failed collection runs make topics look shorter-lived than they were.
• Topics on the list when the range opens, or at the cap boundary, are cut off on the left, so their first_seen is not when they first trended.
• Date filters use UTC; distinct days are counted on the Beijing calendar (UTC+8).
• Rank and heat are the platforms' own and are not comparable across platforms. Observing a topic is not evidence of public opinion, and attention here is not linked to any policy publication.
• The API's list links can open platform searches rather than the original post, and the list is a sample of what the platforms ranked at collection times, not a continuous feed.

Every request behind the figures

These are the exact anonymous requests used. Responses are paged at 100 rows by following next_cursor; the 149 follow-up page URLs, which carry short-lived signed cursors, are listed in the JSON evidence. Keep to one request a second.

# social source names
curl -s "https://china-snapshot.com/api/datasets?dataset=social&limit=1"

# Baidu observations
curl -s "https://china-snapshot.com/api/datasets?dataset=social&source=Baidu%20Top%20Realtime&from=2026-10-04&to=2026-10-31&limit=100"

# Weibo observations
curl -s "https://china-snapshot.com/api/datasets?dataset=social&source=Weibo%20Hot%20Search&from=2026-10-04&to=2026-10-31&limit=100"

# Tencent WeChat observations
curl -s "https://china-snapshot.com/api/datasets?dataset=social&source=Tencent%20WeChat%20Hot%20Topics&from=2026-10-04&to=2026-10-31&limit=100"

# reproducible request: Weibo topics on 2026-10-31 (mode=topics)
curl -s "https://china-snapshot.com/api/datasets?dataset=social&mode=topics&source=Weibo%20Hot%20Search&from=2026-10-31&to=2026-10-31&limit=100"

import time
from urllib.parse import quote

import pandas as pd
import requests

URLS = [
    "https://china-snapshot.com/api/datasets?dataset=social&source=Baidu%20Top%20Realtime&from=2026-10-04&to=2026-10-31&limit=100",  # Baidu observations
    "https://china-snapshot.com/api/datasets?dataset=social&source=Weibo%20Hot%20Search&from=2026-10-04&to=2026-10-31&limit=100",  # Weibo observations
    "https://china-snapshot.com/api/datasets?dataset=social&source=Tencent%20WeChat%20Hot%20Topics&from=2026-10-04&to=2026-10-31&limit=100",  # Tencent WeChat observations
    "https://china-snapshot.com/api/datasets?dataset=social&mode=topics&source=Weibo%20Hot%20Search&from=2026-10-31&to=2026-10-31&limit=100",  # reproducible request: Weibo topics on 2026-10-31 (mode=topics)
]

rows = []
for first_url in URLS:
    url = first_url
    while url:
        page = requests.get(url, timeout=60).json()
        rows += page["items"]
        cursor = page.get("next_cursor")
        url = f"{first_url}&cursor={quote(cursor)}" if cursor else None
        time.sleep(1)  # anonymous limit: 60 requests a minute

df = pd.DataFrame(rows)
if "id" in df: df = df.drop_duplicates("id")
print(len(df), "rows")

# Or load the frozen evidence file without calling the API:
evidence = pd.read_csv("https://china-snapshot.com/data/examples/historical-attention.csv")

Method

1. The reproducible request (Weibo topics on 2026-10-31 (mode=topics)) is one fixed-range page of at most 100 rows; result.row_count and result.ids_sha256 are computed from exactly its response. The figures below come from the wider pull.
2. Read the source names the API reports for dataset=social and match one per platform (Baidu, Weibo, Tencent WeChat).
3. For each platform, page every observation in the range at 100 rows by following next_cursor (capped, see known gaps).
4. Normalize each title: strip a leading rank such as '12. ', collapse whitespace, apply Unicode NFKC. The original-language title is used when present.
5. Group by (platform, normalized topic). first_seen and last_seen are the earliest and latest observation; collection_windows counts distinct collection windows; distinct_days counts Beijing calendar days; best_rank is the lowest rank observed.
6. Median lifespan is last_seen minus first_seen in hours, per platform, with single-window topics counted as zero.
7. Cross-platform overlap is the set intersection of normalized titles. The CSV holds every (platform, topic) row; raw observations are not republished here.