化学PDF数据提取器

快速开始

普通用户可以使用带 bundled_runtime/ 的一键运行包；旧包中的 YiLaiHuanJing/ 作为兼容目录保留：

1. 解压工具包。
2. 双击 Start-Chem-PDF-Extractor.bat。
3. 打开黑色窗口显示的网址。
4. 在网页里填写 API Key。
5. 选择 PDF 文件夹。
6. 配置抽取字段。
7. 点击开始处理。
8. 查看 提取结果.xlsx、错误日志.txt、坏数据.xlsx、可疑数据.xlsx。

更多打包结构、安全检查和发布注意事项见 Windows 一键包说明。

源码用户可以运行：

python -m pip install -r requirements.txt
python -m chem_pdf_extractor

默认情况下，程序只会在终端打印本地 Web UI 地址，不会自动打开浏览器。请复制终端显示的地址到浏览器打开。

如果希望自动打开浏览器，可以运行：

python -m chem_pdf_extractor --open-browser

如果 Windows + Python 3.12 环境下 pymupdf4llm / pymupdf 安装或导入失败，可以使用核心依赖降级路线：

python -m pip install -r requirements-core.txt
python -m chem_pdf_extractor --cli --pdf-mode pypdf_text

Quick Start

git clone https://github.com/myklovenyzforever/chem-pdf-extractor.git
cd chem-pdf-extractor
python -m venv .venv
.\.venv\Scripts\activate
python -m pip install -r requirements.txt
python -m chem_pdf_extractor

By default, the program prints the local Web UI URL in the terminal and does not open the browser automatically. Copy the printed URL into your browser, select a PDF folder, configure extraction fields, and start processing. Test with 3-5 PDFs before running a large batch.

If you want the program to open the browser automatically, run:

python -m chem_pdf_extractor --open-browser

If pymupdf4llm / pymupdf fails to install or import on Windows + Python 3.12, use the core dependency fallback:

python -m pip install -r requirements-core.txt
python -m chem_pdf_extractor --cli --pdf-mode pypdf_text

Alternative script entry:

python run_chem_pdf_extractor.py

The module entry point remains the recommended way to run the project.

Examples

The examples/ directory contains synthetic demonstration files:

• sample_fields.json: chemistry/catalysis-oriented extraction fields.
• sample_output.csv: expected CSV output structure.
• sample_output.xlsx: expected Excel output structure.
• demo_literature_batch/: complete synthetic batch demo with input PDFs, field configuration, and expected output shape.
• demo_literature_batch/：完整合成批处理示例，包含输入 PDF、字段配置和期望输出表格结构。
• field_templates/: reusable field templates for catalysis, materials synthesis, environmental treatment, and electrochemistry workflows.
• field_templates/：面向催化反应、材料合成、环境处理和电化学方向的可复用字段模板。

The example data is synthetic and does not represent real published papers.

Screenshots

A Web UI screenshot and a synthetic Excel/CSV output preview are shown below. Screenshots may vary slightly between releases.

The output preview uses synthetic data only and does not represent real published papers or real extracted research results.

截图

网页界面示例截图和合成 Excel/CSV 输出预览如下，截图可能随版本略有变化。

输出预览仅使用合成数据，不代表真实发表论文或真实抽取结果。

Configuration

This repository does not include any real API key, token, or password.

Normal users can enter the API key in the web interface. The configuration can be saved to config.local.json, which is ignored by Git. config.example.json is only a template and contains placeholders.

Users can enter an OpenAI-compatible Base URL and API key, then use "Fetch models" to load available model IDs when the provider supports the /models endpoint. If model discovery is unavailable, users can still enter the model name manually.

Cloud extraction requires a real API key, Base URL, and model name. Placeholder values are rejected before a task starts.

Environment variables are supported for advanced users, but they are optional and not required for normal use.

$env:CHEM_PDF_EXTRACTOR_API_KEY="YOUR_API_KEY_HERE"
$env:CHEM_PDF_EXTRACTOR_BASE_URL="https://api.example.com/v1"
$env:CHEM_PDF_EXTRACTOR_MODEL="provider/model-name"

配置说明

• 仓库不内置真实 API Key。
• 用户可在网页填写 API Key。
• 可保存到 config.local.json。
• config.local.json 不上传 GitHub。
• config.example.json 只是模板。
• 环境变量只是高级用法，不是必须。
• 本地 Ollama 可作为可选模型后端。
• 使用 OpenAI-compatible API 时，需要自行填写 Base URL、模型名称和 API Key。
• 用户可以填写 OpenAI-compatible Base URL 和 API Key，并在服务商支持 /models 接口时点击“获取模型列表”加载可用模型；如果服务商不支持模型列表接口，也可以手动填写模型名。
• 使用云端抽取时需要填写真实 API Key、Base URL 和模型名；占位配置会在任务开始前被拦截。

One-click Windows Package

The GitHub source repository does not include bundled_runtime/ or YiLaiHuanJing/. A bundled Windows package may be provided through GitHub Releases for non-programming users.

The recommended Windows launcher name is Start-Chem-PDF-Extractor.bat.

For an online first-run Windows release package, users can unzip the package and double-click Start-Chem-PDF-Extractor.bat. The launcher runs install_and_start.ps1, checks for Python 3.11, creates or reuses .venv/, asks the user to choose a PDF backend, installs the matching dependencies, starts the local Web UI, and opens http://127.0.0.1:8766/.

Backend choices:

• pypdf_text: smallest install, fastest installation, best fallback compatibility, weaker layout/table/multi-column handling.
• pymupdf4llm: recommended default, balanced install size and extraction quality, suitable for most research PDFs.
• mineru: optional enhanced backend, larger install size, slower first-time installation, suitable for complex layouts, tables, scanned PDFs, and high-performance PCs. It may require more disk space, memory, installation time, and external downloads.

GitHub Download ZIP is a source package. It does not include Python, .venv/, installed dependencies, MinerU models, or a bundled runtime, but users can still run the first-run launcher online. A fully offline package is not provided by default because it would need bundled Python, wheel caches, MinerU dependencies/models, and larger runtime assets.

Legacy names such as YiJianQiDong.bat and YiLaiHuanJing/ are kept for compatibility with older local packages.

The one-click package may include:

• run_chem_pdf_extractor.py
• Start-Chem-PDF-Extractor.bat
• install_and_start.ps1
• requirements-mineru.txt
• bundled_runtime/
• YiJianQiDong.bat as a legacy launcher
• YiLaiHuanJing/ as a legacy runtime directory

The runtime folder is excluded from the source repository because it is large and machine-specific.

See Windows One-click Package Guide for packaging structure, safety checklist, and release notes.