Doc to JSON

Convert documents (docx, doc, PDF, xlsx, xls) to structured JSON via MinerU. Full pipeline: file to mineru-open-api extract to Markdown then to JSON. Use when user wants to convert a document to JSON, prepare files for local knowledge base RAG, extract structured data from course materials, standards, or reports. Triggers: 转JSON, 转markdown, 知识库素材, doc转json, PDF转JSON, MinerU, 文档解析, extract to json, course standard JSON.

Convert office documents to structured JSON using MinerU as the extraction engine.

Supported Formats

• .doc / .docx — Word documents
• .pdf — PDF files
• .xlsx / .xls — Excel spreadsheets

Prerequisites

• mineru-open-api CLI must be installed (v0.5+)
• MINERU_TOKEN environment variable must be set
• Check: mineru-open-api version

Quick Usage

bash

# Full pipeline: document -> MinerU Markdown -> JSON
python3 scripts/doc_to_json.py /path/to/file.docx -o output.json

# Keep temp files for debugging
python3 scripts/doc_to_json.py /path/to/file.pdf -o out.json --keep-temp

Manual Two-Step Pipeline

If the full pipeline script fails, run steps manually:

Step 1: MinerU Extract

bash

export MINERU_TOKEN="your_token"
mineru-open-api extract input_file.pdf -o /tmp/mineru_out/

Output: .md file in the output directory.

Step 2: Markdown -> JSON

bash

python3 scripts/markdown_to_json.py /tmp/mineru_out/output.md -o output.json

JSON Structure

The output JSON preserves:

• Metadata fields — course name, code, credits, hours, etc. (extracted from plain text)
• Heading hierarchy — 一、二、三... sections become nested keys
• Tables — stored as array of arrays (row cells), keyed as "表格"
• Numbered lists — stored as array of strings under section title
• Paragraph text — merged into "text" field per section

For Knowledge Base Preparation

After JSON conversion, common next steps:

1. Chunk by section — split the JSON into per-section documents for embedding
2. Table extraction — convert "表格" arrays to flattened rows for database import
3. Metadata extraction — pull course code, name, etc. as document metadata
4. Embedding — feed cleaned text chunks into vector database

See references/kb-prep.md for detailed KB preparation patterns.