PDFChatAnnotator

PDFChatAnnotator: A Human-LLM Collaborative Multi-Modal Data Annotation Tool for PDF-Format Catalogs.

🌏 简体中文 / Chinese Version

📝 Description

PDFChatAnnotator is a collaborative annotation tool that leverages the strengths of both human experts and Large Language Models (LLMs) to annotate multi-modal data in PDF-format catalogs. It is designed to streamline and enhance the annotation process through interactive workflows and intelligent suggestions.

📄 Related Publication:

This project is based on our research paper published at ACM IUI 2024:
PDFChatAnnotator: A Human-LLM Collaborative Multi-Modal Data Annotation Tool for PDF-Format Catalogs

📌 Version Update

Current Version: 2.0

• In version 1.0, data was saved to a MySQL database, which required additional setup and configuration.
• In version 2.0, to simplify the installation and usage process—especially for non-computer science users—we have switched to saving annotation results directly into Excel files (.xlsx format).
  This change makes the tool more accessible and easier to use out of the box.

📊 System Overview

🖍️ Interactive Annotation Interface

📌 Before You Start

• The currently supported catalog types are:

  • Each page's images are only associated with the text content on that page (a).

  • All images that appear before the start of new page text are associated with the text content on the current page (b).

  • In a page where there are multiple image-text matching pairs, each image is associated with the text content below it (c). ⚠️ Due to the high correlation with the inherent characteristics of the catalog type, it is currently not open source.

    

⚙️ Installation

Prerequisites

• Python 3.9
• Anaconda (recommended for environment management)
• Visual Studio Code (recommended IDE)

1. Download and Set Up the Project

1. Download the project:

   • Visit: https://github.com/VanillaTY/PDFChatAnnotator
   • Click the green Code button and select Download ZIP
   • Extract the ZIP file to your preferred location (e.g., Desktop)

2. Open the project in VS Code:

   • Drag the extracted folder into VS Code
   • If prompted with "Do you trust the authors?", select "Yes"

2. Set Up Python Environment

Using Anaconda (Recommended)

1. Install Anaconda:

   • Download from: https://www.anaconda.com/download
   • Follow the installation wizard
   • For Windows users: Add Anaconda to system PATH during installation

2. Create and activate the environment:

   conda create -n pdfannotator python=3.9
   conda activate pdfannotator

3. Install Dependencies

1. Install project dependencies:

   pip install -r requirements.txt

2. Install OS-specific dependencies:

   • Windows:

     pip install pyreadline3

   • macOS:

     pip install readline

4. Configure API Key

1. Obtain your API key:

   • Visit: https://api.chatanywhere.tech/#/
   • Purchase a plan and get your API key

2. Configure the API key:

   • Open utils/prompt.py
   • Replace the placeholder with your API key:

     api_key = "your_api_key_here"
     base_url = "your_base_url_here"

🚀 Running the Application

1. Activate the environment:

   conda activate pdfannotator

2. Start the development server:

   python manage.py runserver

3. Access the application:

   • Open your browser
   • Navigate to: http://127.0.0.1:8000/

📄 PDF Preprocessing (Required Before Use)

Before launching the system, you must preprocess your PDF file(s) to extract necessary text and image data.

Please follow the guide below before running the application:

• 📘 中文预处理教程
• 📙 English Preprocessing Guide

The preprocessing process requires a GPU-supported environment and will prepare the data required for annotation.

📌 Quick Start Guide

For daily use:

1. Open VS Code and load the project
2. Open terminal and run:

   conda activate pdfannotator
   python manage.py runserver

3. Access http://127.0.0.1:8000/ in your browser

For more detailed installation instructions, please refer to the Installation Guide.