Bilibili Danmaku Scraper

Measure timed audience reactions with bilibili danmaku text samples and observed-pool profiles. Supply public BV IDs or video URLs and get one structured record per video part: timed texts, minute counts, peak minute, pool markers and aggregate unique-sender counts. No login cookies are needed. This is a snapshot of Bilibili's public XML pool, not a complete comment archive.

Who is it for?

Video researchers can locate moments with concentrated audience reactions. Editors can inspect sampled texts near a reaction peak. China-market analysts can compare public reaction pools across supplied videos. Data teams can join snapshots to video metadata by BV ID and content ID.

Why use this Actor?

You receive a compact video-part profile instead of thousands of independently billed comment rows. Metrics use the whole observed pool even when you request few text samples. Sender hashes are counted only in memory and never exported. The output distinguishes lifetime counters from fetched pool counts rather than presenting a sample as full coverage.

Getting started

1. Choose a public Bilibili video and copy its BV ID or canonical HTTPS video URL.
2. Enter it in videos. Add ?p=2 to a URL to select part 2 of a multipart video.
3. Choose sampleSize (0–500) and maxItems (1–100).
4. Run the Actor and inspect the default dataset.
5. Export JSON for nested samples or flatten the arrays in your downstream pipeline.

Example input:

{
  "videos": ["BV1xx411c7mD"],
  "maxItems": 10,
  "sampleSize": 100
}

Inputs

Field	Default	Behavior
videos	Required	1–100 public BV IDs or HTTPS bilibili.com/video/BV… URLs. IDs select part 1. URL p selects a part.
maxItems	10	Global maximum unique BV/part profiles in input order, after deduplication. Range 1–100; zero/unlimited are invalid.
sampleSize	100	Maximum timed texts per profile, 0–500. Zero returns metrics with an empty samples array.

Duplicate BV/part pairs are removed before applying maxItems. There is no keyword search, discovery, short-link resolution, av-ID input, cookie input or hidden filter. Unsupported fields and malformed values fail validation. Only supplied video parts are collected; a multipart video is not automatically expanded.

Bilibili danmaku sampling and coverage

The Actor resolves the selected part's content ID using public metadata, then reads its public XML pool. Entries are sorted by playback position. Text samples are selected at evenly spread indices in that sorted sequence, including both endpoints when at least two samples are requested. This is not random sampling or uniform spacing in seconds.

Minute buckets, sender count and playback bounds use every observed entry, independently of sampleSize. Missing minute buckets mean zero comments in this observed pool. The highest-count minute wins; ties choose the earliest minute. The overall rate divides observed count by the selected part's reported duration in minutes, not by the sampled time span.

poolLimit is the source's maxlimit marker. Actual fetched entries can exceed it. poolCapped is true when fetched count reaches that marker, or is below a known single-part lifetime total. A false marker does not prove full history. A zero pool can be legitimate and still produces a video-part profile.

totalDanmaku is the lifetime video counter only for single-part videos. It is null for multipart videos, because a video-wide count cannot honestly be assigned to one selected part. Sender counts are distinct source hashes in the observed pool, not verified unique people.

Extracted data

Field	Meaning
bvid, videoUrl, page, cid	Video identity, selected part and public content ID
title, partTitle	Video title and optional part label
danmakuFetched	Number of entries in the observed public XML pool
uniqueSenders	Aggregate distinct hashes in that pool; hashes themselves are excluded
poolLimit, poolCapped	Source pool marker and conservative coverage indicator
totalDanmaku	Single-part lifetime video counter, otherwise null
perMinute	Nonzero {minute, count} playback buckets
peakMinute	Highest observed bucket, or null for an empty pool
firstTimeSeconds, lastTimeSeconds	Observed playback bounds, null when empty
durationSeconds, danmakuPerMinute	Selected part duration and observed count per duration-minute
samples, sampleCount	Timed text samples and their count
coverage, scrapedAt	Coverage declaration and UTC snapshot timestamp

Each sample contains timeSeconds, text, mode, fontSize, decimal RGB color and UTC sentAt. No sender ID, hash or profile is returned.

Output example

A local snapshot of BV1xx411c7mD returned 3,600 pool entries with a 3,000 pool marker. The shortened example below shows the actual shape and aggregate values; text is anonymized and arrays are abbreviated, not the full export.

{
  "bvid": "BV1xx411c7mD",
  "videoUrl": "https://www.bilibili.com/video/BV1xx411c7mD/?p=1",
  "page": 1,
  "cid": 62131,
  "title": "Example video",
  "partTitle": "",
  "danmakuFetched": 3600,
  "uniqueSenders": 2951,
  "poolLimit": 3000,
  "poolCapped": true,
  "totalDanmaku": 128639,
  "perMinute": [{"minute": 0, "count": 3280}],
  "peakMinute": {"minute": 0, "count": 3280},
  "firstTimeSeconds": 0,
  "lastTimeSeconds": 2054.425,
  "durationSeconds": 2055,
  "danmakuPerMinute": 105.1094890510949,
  "samples": [{"timeSeconds": 0, "text": "Example reaction", "mode": 5, "fontSize": 25, "color": 15138834, "sentAt": "2020-04-23T04:22:43.000Z"}],
  "sampleCount": 100,
  "coverage": "observed-public-xml-pool",
  "scrapedAt": "2026-10-08T20:07:44.376Z"
}

How much does it cost to sample Bilibili danmaku?

A run has a $0.005 start event and one pool-profile event per saved video part. Timed texts and minute buckets are included in that profile; they have no separate event charge. The profile price is $0.00884 on BRONZE, $0.010166 on FREE, $0.0068952 on SILVER and $0.005304 on GOLD, PLATINUM and DIAMOND. Spend tiers follow your qualifying aggregate monthly Store spend, not this Actor's individual run volume.

At BRONZE, total run event charges including start are $0.01384 for one saved profile, $0.0934 for ten and $0.889 for 100. FREE estimate: $0.015166 total for one run with one saved profile. These examples concern Actor event charges, not any separate account/platform charges. A valid empty-pool profile is still a profile. Failed requests do not generate profile events; the start event and any already saved profiles remain charged.

Limits and failure behavior

Public XML pools change over time and may omit historical comments, suppressed texts or restricted videos. This Actor does not download media, subtitles, comment replies, historical segments or sender profiles. Snapshot differences are not guaranteed new-comment feeds, deletions or population trends.

Requests are sequential with a 20-second timeout. Transient network errors, HTTP 429 and 5xx receive at most two retries. Stable 4xx/challenges fail without blind retries. No automatic paid proxy or browser fallback is enabled. Metadata/XML errors fail the run; previously saved rows are partial, not a completed full result.

Each upstream response is capped at 4 MiB and each XML pool at 50,000 entries. Oversized pools fail visibly rather than being silently truncated. Default run timeout is five minutes; choose a smaller input batch when the source is slow. A billing limit can also stop collection; check run status and returned counts.

Integrations

Use JSON exports for nested samples. Unwind samples into spreadsheet rows when individual texts are needed; unwind perMinute for reaction charts. Join BV/part keys with Bilibili Scraper for video metadata and top-level comments.

An Apify schedule can repeat the same supplied-video snapshot. Save snapshot times and compare observed counts downstream; do not interpret a changing pool as a complete archive. Webhooks can notify your ETL pipeline when a run succeeds. Make, Zapier and n8n can retrieve the default dataset through Apify's normal integrations.

API usage

Start a run with your own authorized Apify token:

curl -X POST 'https://api.apify.com/v2/acts/automation-lab~bilibili-timed-danmaku-samples/runs' \
  -H "Authorization: Bearer $APIFY_TOKEN" \
  -H 'Content-Type: application/json' \
  -d '{"videos":["BV1xx411c7mD"],"sampleSize":100}'

JavaScript:

import { ApifyClient } from 'apify-client';
const client = new ApifyClient({ token: process.env.APIFY_TOKEN });
const run = await client.actor('automation-lab/bilibili-timed-danmaku-samples')
  .call({ videos: ['BV1xx411c7mD'], sampleSize: 100 });
if (run.status !== 'SUCCEEDED') throw new Error(`Run ${run.id}: ${run.status}`);
const { items } = await client.dataset(run.defaultDatasetId).listItems({ limit: 10 });
console.log(items);

Python:

import os
from apify_client import ApifyClient
client = ApifyClient(os.environ['APIFY_TOKEN'])
run = client.actor('automation-lab/bilibili-timed-danmaku-samples').call(
    run_input={'videos': ['BV1xx411c7mD'], 'sampleSize': 100})
if run['status'] != 'SUCCEEDED':
    raise RuntimeError(f"Run {run['id']}: {run['status']}")
print(client.dataset(run['defaultDatasetId']).list_items(limit=10).items)

MCP usage

Use this Actor-specific selection with an authorized Apify MCP client:

claude mcp add --transport http apify \
  'https://mcp.apify.com?tools=automation-lab/bilibili-timed-danmaku-samples'

Claude Desktop, Cursor and VS Code: use the equivalent client configuration below (or your client's HTTP MCP configuration UI).

{"mcpServers":{"apify":{"url":"https://mcp.apify.com?tools=automation-lab/bilibili-timed-danmaku-samples"}}}

Discover actual names and argument schemas with tools/list. Actor selection also adds run/dataset/KV retrieval and abort tools; discovery is not permission to run or abort anything. Example prompt (authorized invocation): “Sample 100 timed danmaku texts from BV1xx411c7mD and report its observed peak minute, clearly stating pool coverage.”

Start once and retain the run ID, status and dataset ID; IDs are not result rows. call-actor waits at most 45 seconds (default 30). If still running, check the same run with 2/4/8-second backoff capped at 10 seconds and a 120-second total deadline. Set the client timeout above the chosen wait plus transport margin. On client timeout recover the known run, not a new run; if its ID is unknown report uncertainty. At deadline report pending and stop.

After success, read bounded dataset pages with explicit limit, offset and required fields. Example consumer budget: 20 source rows per page, at most 100 rows or 64 KiB serialized UTF-8 across admitted pages, whichever comes first. Enforce bytes host-side before model injection; omit or summarize an oversized row with disclosure. If your client cannot intercept oversized responses, do not claim a hard byte guarantee. Track source offsets, continuation and partial/truncated status; keep full exports outside model context. Discovery caching does not guarantee token savings.

Legality, responsible use and diagnostics

Use public content only for a lawful purpose, respect Bilibili's terms, content rights and applicable privacy laws, and avoid re-identifying commenters. Public availability is not a legal license. Texts can contain personal information even though this Actor removes structured sender identifiers. Apply downstream retention and access controls appropriate to your use.

This independent Actor is not affiliated with or endorsed by Bilibili. It uses no AI model at runtime; text is source content, not generated sentiment or summaries. Results use your Apify storage retention settings; delete datasets and runs through Apify controls. There is no persistent private comment cache. For help, use the Actor's Apify Issues tab and include a safe reproduction input and run link.

Failed operations send sanitized diagnostic input, exceptions and Actor/build/run IDs to our private GlitchTip service for repair. Secret fields and URL queries are removed, and reports are retained for 30 days. No diagnostic endpoint or credential is part of the public input.

FAQ and troubleshooting

Is this all danmaku ever sent? No. It is the current observed public XML pool, with bounded text samples.

Why does fetched count exceed poolLimit? Bilibili can return more entries than its advertised marker. We report both instead of silently clamping the data.

Why do counts differ across runs? Pools and lifetime counters change independently. Compare scrapedAt, selected part and coverage before interpreting differences.

Why is totalDanmaku null? The selected video has multiple parts; the video-wide total cannot be assigned to a single part.

What if I see 412 or another upstream failure? Check video availability and run logs. The run fails honestly rather than exporting an error as a profile. Reduce your batch and try later; no login or paid recovery is silently attempted.

Can I request no texts? Set sampleSize to zero. Pool metrics remain available and the profile event still applies.

Related Actor

Bilibili Scraper complements these timed pool profiles with public video metadata, engagement and optional top-level comments. Neither product promises complete comment history.