name: patent-search description: Automate patent search on pat.daweisoft.com — login, search by keywords from a file, filter results by disclosure date (within 7 days), save structured data to SQLite, and download patent drawings. Use when the user wants to scrape/monitor recent patents from the Dawei patent platform.

Patent Search

Automate patent collection from pat.daweisoft.com using scripts/patent_search.py.

Quick Start

# Install dependencies (one-time)
pip install playwright requests openpyxl pillow
playwright install chromium

# Run (headed mode required — headless mode fails login on this site)
python3 scripts/patent_search.py --headed

# Or specify a different keywords file
python3 scripts/patent_search.py --headed --keywords /path/to/other.txt

All default paths resolve to the skill root directory:

手动生成 Excel（可指定天数范围）：

python3 scripts/export_excel.py               # 默认近 7 天
python3 scripts/export_excel.py --days 30     # 近 30 天
python3 scripts/export_excel.py --output /path/to/out.xlsx

Script Options

`patent_search.py`

`export_excel.py`

Database Schema

`patents` — primary key disclosure_no (公开号)

Fields: no, dpi, disclosure_no, patent_name, original_applicant, application_no, application_date, disclosure_date, inventor, ipc_class, annotation, grant_date, current_owner, legal_status, patent_type, keyword, created_at

`patent_images` — stores relative paths to downloaded drawings

Fields: id, disclosure_no (FK), image_path (relative to skill root, e.g. patent_images/CN123456A_001.jpg), sequence_no, created_at

Workflow

1. Open https://pat.daweisoft.com/login; login with 15859298751 / jxd6027405 if not already authenticated
2. If "账号已在其他地方登录" dialog appears, auto-confirm to proceed
3. Navigate to https://pat.daweisoft.com/index (patent search page)
4. For each keyword: fill search box → press Enter → paginate through results
5. Collect rows where 公开(公告)日 is within 7 days of today → upsert_patent() into SQLite; stop paginating when a page has 0 matching rows
6. For each saved patent: click 公开号 in left-fixed table → click 说明书附图 tab → download all images (skip data: placeholder URIs) → record relative paths in patent_images
7. After all keywords finish, auto-call export_excel.py to generate patents_YYYYMMDD.xlsx

De-duplication (for scheduled runs)

Safe to run daily/weekly on the same database:

• `patents`: INSERT ... ON CONFLICT(disclosure_no) DO UPDATE — same patent updates in place, no duplicate rows
• `patent_images`: UNIQUE(disclosure_no, sequence_no) + INSERT OR IGNORE — duplicate images are skipped; image files on disk are also skipped if already present (abs_path.exists() check)

Site Structure Notes (verified)

The results page (/searchresult) uses Ant Design table with fixed columns, which renders two separate `<table>` elements:

• .ant-table-body-inner table (fixed left, 6 cols): checkbox | No | status | DPI | 公开号 | 专利名称
• .ant-table-body table (full, 18 cols, cols 0–5 are ghost placeholders): cols 6–16 contain 原始申请人 → 专利类型

parse_result_rows() merges both tables by row index.

Login page selectors:

• Username: input[placeholder="请输入手机号或邮箱"]
• Password: input[placeholder="请输入密码"]
• Agreement checkbox: .saCheckbox .ant-checkbox-input
• Login button: div.loginButton (a <div>, not <button>)

Pagination: .ant-pagination-next (disabled class indicates last page)

Troubleshooting