RGPRec

This repository contains the implementation of RGPRec, a RAG‐enhanced GNN for personalized task recommendations in open-source communities such as SourceForge and GitHub.

Model Overview

The RGPRec framework consists of three main components:

• RAG-enhanced Interactive Data Augmentation: Retrieves relevant project documents and uses LLMs to generate developer/task attributes.

• Developer Ability Rating: Employs multi-label neural networks to predict each developer's rating for tasks based on historical behavior and semantic features.

• Personalized Recommendation: Constructs ego graphs for each developer to perform localized reasoning and generate debiased rating scores.

Main Results

Project Structure

RGPRec/
│
├── RGPRec-RAG/          # Raw data and RAG‐enhanced interactive data augmentation
├── RGPRec-Ability/      # Developer ability rating
├── RGPRec-Ego/          # Personalized recommendation and evaluation
└── requirements.txt     # Python dependencies

Project Setup

Environment Setup

Please use the following commands to create and activate the environment:

conda create -n RGPRec python=3.6
conda activate RGPRec
pip install -r requirements.txt

Quick Test

To quickly test RGPRec without training, simply run:

cd RGPRec-Ego
python main-rank.py

Train from Scratch

If you'd like to retrain the model or reproduce our experimental results, follow these steps:

Step 1: Data Preparation

• The raw data of SourceForge and GitHub is provided in the RGPRec-RAG/Tensorflow_raw/ and RGPRec-RAG/SourthForge_raw/ folders.
• Please configure the APIs for GPT-4o and text-embedding-ada-002, and obtain the Developer and task Attributes through Cherry Studio(https://github.com/CherryHQ/cherry-studio).

Step 2: Ability Rating

• Run ability rating using:

cd RGPRec-Ability
python RGPRecA.py

This will generate ability scores and save them.

Step 3: Recommendation Model Training & Evaluation

• Train and evaluate the main recommendation model:

cd ../RGPRec-Ego
python main.py --dataset_path ../RGPRec-Ability/dataA_result/Tensorflow/ --data ../RGPRec-Ability/dataA_result/Tensorflow

• For ranking-based evaluation:

python main-rank.py --dataset_path ../RGPRec-Ability/dataA_result/Tensorflow/ --data ../RGPRec-Ability/dataA_result/Tensorflow