Bilibili Video Scraper

Scrape Bilibili — China's largest video platform — via its public api.bilibili.com endpoints (no login). Search videos by keyword, fetch single video details by bvid / aid / URL, get the popular feed, or partition rankings. Output includes title, author, view counts, duration, cover image, publish time and tags.

What this actor does

• Four modes: search (default), byVideoUrl, popular, ranking
• Search ordering (order): total rank / most viewed / newest / most danmaku / rating score
• 15 ranking partitions as a dropdown (rid): all, anime, music, game, entertainment, knowledge, guichu, dance, fashion, life, movie, digital, food, animal, car — every option verified against the live ranking API
• Search partition filter (tids) and search duration filter (duration: <10 / 10-30 / 30-60 / >60 min), both verified against the live search API
• buvid bootstrap: visits bilibili.com once to obtain the buvid3 cookie the search API requires
• Web-search HTML fallback (search.bilibili.com SSR cards, with full title/author/views/duration/cover parsing) if the search API rate-limits
• Filters: min view count, title/description/author keyword
• Auto-escalation: lazy-engages Apify AUTO datacenter proxy on 403/429/5xx; retries with backoff (bilibili is latency-flaky)
• Empty fields are omitted; every record carries sourceUrl, scrapedAt, recordType

Output per video

• bvid, aid
• title (HTML <em> keywords stripped), description
• author, authorMid
• viewCount, danmaku, replyCount, favorites, likeCount, coinCount, shareCount
• duration (seconds), durationText (3:17:23)
• coverUrl (hdslb CDN, https-normalized), publishTime (UTC ISO)
• typeName (partition label), uploadRegion (upload location, when published), pageCount (multi-part videos), resolution (1920x1080)
• tags[] (search mode)
• publishDate (YYYY-MM-DD, HTML-fallback records only; relative SSR labels like "5小时前" are dropped — only real dates are emitted)
• videoUrl, recordType: "video" | "error", scrapedAt

Input

Field	Type	Default	Description
mode	select	search	search / byVideoUrl / popular / ranking
searchQuery	string	python	Keyword (mode=search)
videoUrls	array	–	Video URLs (mode=byVideoUrl)
bvids	array	–	BV IDs (mode=byVideoUrl)
aids	array	–	Numeric video IDs (mode=byVideoUrl)
rid	select	0	Ranking partition (15 options)
order	select	totalrank	Search sort
tids	select	0	Search partition filter
duration	select	0	Search duration range
minViewCount	int	–	View-count floor
containsKeyword	string	–	Title/description/author substring filter
maxItems	int	20	Hard cap (1–500)

Example: most-viewed Python tutorials

{
  "mode": "search",
  "searchQuery": "python",
  "order": "click",
  "minViewCount": 100000,
  "maxItems": 15
}

Example: 30-60 minute food videos from the food partition

{
  "mode": "search",
  "searchQuery": "美食",
  "tids": "211",
  "duration": "3",
  "maxItems": 10
}

Example: single video

{
  "mode": "byVideoUrl",
  "videoUrls": ["https://www.bilibili.com/video/BV1wD4y1o7AS/"]
}

Example: music ranking

{
  "mode": "ranking",
  "rid": "3",
  "maxItems": 20
}

Cover images

Cover URLs point at i*.hdslb.com — verified HTTP 200 image/jpeg from a clean shell (no Referer/cookies); protocol-relative and http:// URLs are normalized to https://.

Anti-bot notes

• api.bilibili.com search endpoints require a buvid3 cookie — the actor bootstraps it by visiting bilibili.com first.
• The platform rate-limits aggressively (-799/-1200 codes, intermittent timeouts) — retries with backoff are built in; web-search HTML is the fallback.

Data source

Public api.bilibili.com endpoints: x/web-interface/search/type, x/web-interface/view, x/web-interface/popular, x/web-interface/ranking/v2. No login required.

Limitations

• User-space video lists (x/space/wbi/arc/search) require WBI request signing — not supported in v1.
• The 资讯 (news) and 运动 (sports) partitions and the invalid tid 153 were removed from the rid/tids dropdowns because the ranking/search APIs reject them (-400).
• Search-mode tids and duration filters apply to the API path; if the search API rate-limits or times out mid-run (common — the platform is latency-flaky), the HTML fallback returns unfiltered results, so the filter may not hold on every run. Retrying the run usually returns the filtered set.
• Garbage search queries: the platform returns real recommended videos instead of an empty result set, so an "impossible" query still yields real records (platform behavior, not a filter leak).
• Invalid bvids/URLs produce typed recordType: "error" records (no silent drops).