| name | prtimes |
| description | Platform rules, internal API discovery, Taiwan-vs-Japan filter, and date extraction for the prtimes scraper |
| applyTo | scraper/sources/prtimes.py |
PR TIMES Scraper — Platform Reference
PR TIMES (prtimes.jp) is Japan's largest press release distribution platform.
Taiwan-related event announcements are submitted by event organisers, venues, and
tourism/trade bodies and are a reliable early-warning source for upcoming Japan–Taiwan
cultural events.
Platform Profile
| Field | Value |
|---|
| Site URL | https://prtimes.jp |
| API/Rendering | Internal JSON API (server-rendered detail pages) |
| Auth required | No — User-Agent header only |
| Rate limit | Not documented; 0.5 s polite delay is sufficient |
| Source name | prtimes |
| Source ID format | prtimes_{release_id} (e.g. prtimes_68) |
API Discovery
The search endpoint was found by analysing the Next.js bundle
(_app-d5e27ce51595715c.js, module 20400):
G = `${$}/keyword_search.php` where $ = "${location.origin}/api"
→ https://prtimes.jp/api/keyword_search.php
Search endpoint:
GET https://prtimes.jp/api/keyword_search.php/search
?keyword=<url-encoded>&page=<N>&limit=40
Response:
{
"data": {
"current_page": 1,
"last_page": 12,
"limit": 40,
"release_list": [
{
"release_id": 12345,
"title": "...",
"company_name": "...",
"released_at": "2026年4月25日 17時00分",
"release_url": "/main/html/rd/p/000012345.html"
}
]
}
}
Results are sorted newest-first. Maximum 250 pages × 40 = 10,000 results per keyword.
Field Mappings
| Event Field | Source |
|---|
source_id | prtimes_{release_id} |
source_url | https://prtimes.jp + release_url |
raw_title | title from API response |
raw_description | `"開催日時: YYYY年MM月DD日\n\n" + detail-page body text (first 3000 chars) |
start_date | Extracted from detail-page body (see Date Extraction) |
end_date | Same as start_date (single-day default) |
location_name | Extracted from 会場： / 開催場所： label in body |
location_address | Same as location_name (not split further) |
original_language | "ja" |
Taiwan Relevance Filter
Two-stage title filter applied before fetching detail pages:
- Taiwan keyword:
台湾|Taiwan|台灣|臺灣 must appear in title.
- Event-type keyword:
イベント|フェス|開催|展示|祭|セミナー|講演|体験|交流会|… must appear in title.
Taiwan-based event exclusion (applied after detail fetch):
- Title pattern
_TAIWAN_BASED_TITLE_RE: matches in 台湾, in Taiwan, 台湾進出, 台湾販路, 台湾.*に集結 etc.
- Venue pattern
_TAIWAN_VENUE_RE: venue contains 台北, 台中, 高雄, 新竹, 花蓮, Taipei, Kaohsiung etc.
Noise estimate: ~2.8% of results are Japan-held Taiwan events. The annotator
(annotator.py) acts as a second-pass filter.
Date Extraction Notes
Priority order in _extract_date_from_body():
- Labeled pattern (highest confidence):
開催日時： / 日時： / 開催日： / 開催期間： → extract YYYY年MM月DD日 HH:MM if present
- Year-less labeled pattern:
M月D日 after the label → inject year from released_at, advance 1 year if already past
- Standalone date scan:
All YYYY年MM月DD日 occurrences in body — skip index 0 (PR dateline), use first plausible candidate (not older than 2 years)
- Fallback:
released_at (PR publication date)
Sanity check: dates older than 2 years from now are rejected and replaced with released_at.
Troubleshooting
| Symptom | Likely Cause | Fix |
|---|
| 0 results for a keyword | API changed path | Check JS bundle for new keyword_search endpoint |
start_date equals PR publish date | Labeled date pattern absent from detail page | Check HTML selector in _fetch_detail; body may be in different container |
| Taiwan-based events not filtered | Title pattern doesn't match | Add matching pattern to _TAIWAN_BASED_TITLE_RE or _TAIWAN_VENUE_RE |
| Duplicate events | Same PR submitted under multiple keywords | seen_ids set deduplicates by release_id within a run; Supabase upsert handles cross-run dedup |
| HTTP 429 / rate limit | Too many requests | Increase _DELAY (currently 0.5 s) |
| Detail page returns 404 | PR was deleted | _fetch_detail returns ("", "") gracefully; raw_description falls back to title |
Pending Rules
