DKG: ORCDF-based Cognitive Diagnosis Engine

This project provides a backend engine for student cognitive diagnosis based on the ORCDF (Oversmoothing-Resistant Cognitive Diagnosis Framework) model. It leverages Graph Neural Networks (GNNs) to predict student performance on exercises, offering a powerful API for real-time diagnostic predictions.

本项目是一个基于 ORCDF (抗过平滑认知诊断框架) 模型的学生认知诊断后端引擎。它利用图神经网络 (GNN) 预测学生的习题表现，并提供一个强大的API进行实时诊断预测。

---

System Architecture / 系统架构

graph TD
    subgraph "数据层 (Data Layer)"
        A["原始CSV文件 <br> student_logs.csv"]
    end

    subgraph "处理与训练层 (Processing & Training Layer)"
        B["data_loader.py <br> (加载 & 预处理)"]
        C["ORCDF Model <br> (orcdf/model.py)"]
        D["Trainer <br> (train_orcdf.py)"]
        E["持久化模型 <br> orcdf_model.pt"]
    end

    subgraph "应用层 (Application Layer)"
        F["API 服务器 <br> api_server.py"]
        G["教育应用 / 第三方服务"]
    end

    A --> B;
    B --> D;
    C --> D;
    D -- "训练/保存" --> E;
    E -- "启动时加载" --> F;
    F -- "提供端点" --> G;

Loading

✨ Core Features / 核心功能

• High-Precision Prediction: Utilizes the advanced ORCDF GNN model to accurately predict student performance.
• Oversmoothing Resistance: Employs a specialized architecture and consistency regularization to mitigate the oversmoothing issue common in GNNs, ensuring more distinct student representations.
• Real-time API: Offers a high-performance FastAPI backend to serve real-time prediction requests.
• Scalable Framework: Built on PyTorch, providing a solid foundation for future extensions and research.

---

🚀 Quick Start / 快速开始

1. Clone Repository:

   git clone https://github.com/MuQY1818/DKG.git
   cd DKG

2. Install Dependencies:

   pip install -r dkg_mvp/requirements.txt

3. Train the GNN Model: Before the first run, you must train the ORCDF model and generate the model file.

   # This command will train the model on the full dataset
   # The best model will be saved in the 'models/' directory
   python -m dkg_mvp.train_orcdf

4. Run API Server:

   python api_server.py

   Once the server is running, visit http://127.0.0.1:5000/docs in your browser to see the interactive API documentation for the new prediction endpoints.

---

For More Information / 详细信息

For detailed documentation, please choose your preferred language: 要获取更详细的文档，请选择您的语言：

• English
• 简体中文
• Technical Documentation (English)
• 技术文档 (中文)