Social Media Liked Collector

Overview

Use the local social_zongjie Node project to collect original liked-content text from Douyin and Xiaohongshu into Excel. The tool connects to a dedicated Edge session over CDP, preserves source links, uses local Tesseract OCR for image text, and does not call external AI services.

Project path:

C:\Users\Administrator\Desktop\code\social_zongjie

Required Boundaries

• Do not bypass login, CAPTCHA, permission prompts, rate limits, or platform risk-control screens. Stop and hand off clearly.
• Do not like, favorite, comment, follow, post, message, upload, or change account settings.
• Do not inspect cookies, local storage, passwords, session stores, or hidden private data.
• Use browser-visible content and the project's generated rows as the source of truth.
• Keep OCR local. The default Tesseract paths are D:\Tools\Tesseract-OCR\tesseract.exe and D:\Tools\Tesseract-OCR\tessdata.

Workflow

1. Work from the project directory:

Set-Location 'C:\Users\Administrator\Desktop\code\social_zongjie'

2. Verify dependencies are present. If node_modules is missing, run npm install. The project postinstall links Codex's spreadsheet artifact tool.

3. Ensure .env exists. If missing:

Copy-Item .env.example .env

4. Verify local OCR before collecting image-text notes:

D:\Tools\Tesseract-OCR\tesseract.exe --version
D:\Tools\Tesseract-OCR\tesseract.exe --list-langs

--list-langs must include chi_sim, eng, and preferably osd. If Chinese language data is missing, stop and report it.

5. Start the dedicated Edge session if it is not already running:

powershell -ExecutionPolicy Bypass -File scripts\start-edge.ps1

Ask the user to log in manually if Douyin or Xiaohongshu is not logged in. Do not close this Edge window while collecting.

6. Run collection:

npm start

The default processes 20 Douyin items and 20 Xiaohongshu items.

Use targeted commands when the user specifies scope:

node src\cli.mjs --platform both --douyin-limit 10 --xhs-limit 15
node src\cli.mjs --platform douyin --douyin-limit 20
node src\cli.mjs --platform xhs --xhs-limit all
node src\cli.mjs --platform both --resume

7. Report the generated Excel path printed as Excel 已生成: .... Outputs are written under:

C:\Users\Administrator\Desktop\code\social_zongjie\output

What The Tool Collects

• Douyin liked videos: opens content, uses the platform's built-in AI entry when available, asks for 视频总结, and saves the original answer.
• Douyin image-text posts: saves visible copy and local OCR text from post images.
• Xiaohongshu liked image-text notes: saves visible note text and local OCR text from all reachable images.
• Excel workbook sheets include Douyin content, Xiaohongshu content, and run logs. Rows retain clickable source links.

Resume And Recovery

• output\state.jsonl stores processed rows for resume support.
• Use --resume after interruption to skip already processed content IDs.
• A per-item failure should be recorded in output and collection should continue.
• A platform-level failure such as login loss, CAPTCHA, blocked access, or page redesign should stop that platform and be included in the final handoff.

Common Checks

• If Edge cannot connect, confirm the dedicated browser was launched with remote debugging on the .env EDGE_DEBUG_URL, default http://127.0.0.1:9222.
• If output export fails, run npm install again to refresh the artifact-tool link, then run npm test -- test\artifact-link.test.mjs if needed.
• If OCR is empty or garbled, confirm Tesseract language data, then trust the project's preprocessing/retry path. Mark uncertainty instead of inventing text.
• If Xiaohongshu needs an unbounded liked traversal, prefer --xhs-limit all; Douyin requires a positive integer limit.

Verification

Before claiming completion, run the narrow tests that do not require live platform access:

npm test -- test\core.test.mjs test\processing.test.mjs test\excel.test.mjs test\artifact-link.test.mjs

For live collection, successful verification is the printed Excel path plus a workbook under output\ containing the requested platform rows or clearly logged platform errors.