Entity Alignment Using RAG (EA_RAG)

Overview

This repository introduces EA_RAG, a novel framework for entity alignment in knowledge graphs (KGs) using Retrieval-Augmented Generation (RAG) with Large Language Models (LLMs). The framework dynamically integrates retrieval mechanisms with generative models to align entities across multilingual and multi-modal KGs.

The primary contribution of this project is the implementation of RAG-based alignment methods, evaluated against traditional embedding-based approaches provided by the EAkit framework.

Features

• RAG-Based Alignment: A novel entity alignment approach leveraging generative capabilities of LLMs.
• Dynamic Retrieval: Incorporates external knowledge during the alignment process for improved accuracy.
• Evaluation Using EAkit: Benchmarked against state-of-the-art embedding-based methods such as GCN-Align, MTransE, and BootEA.
• Cross-Lingual Matching: Optimized for datasets like DBP15K, supporting multilingual knowledge graphs.
• Scalable Architecture: Uses Pinecone for efficient vector storage and retrieval.

---

Prerequisites

General Requirements

• Python: 3.11
• Conda: Miniconda or Anaconda (Download Miniconda)
• Git: Installed and configured (Download Git)
• Pinecone: API for vector embedding management (Pinecone Documentation)

---

Getting Started

Clone the Repository

Clone the repository with submodules:

git clone --recurse-submodules https://github.com/DariMe20/Entity-Alignment-KG
cd Entity-Alignment-KG

Create a Conda Environment for EA_RAG

1. Create and activate the environment:

   conda create --name ea_rag python=3.11 -y
   conda activate ea_rag

2. Verify the Python version:

   python --version  # Should output 3.11.x

Install Dependencies

Install required packages using pip:

pip install -r requirements.txt

Configure API Keys

Create a .env file with the following details:

OPENAI_API_KEY01=your_openai_key
OPENAI_API_KEY02=your_openai_key
OPENAI_API_KEY03=your_openai_key
OPENAI_API_KEY04=your_openai_key
OPENAI_API_KEY05=your_openai_key
PINECONE_API_KEY=your_pinecone_key

---

Dataset Setup

Download DBP15K Dataset

1. Download the DBP15K dataset from Hugging Face.
2. Extract and rename the folder to fr_en in the repository directory.

Using EA_RAG

The implementation and usage of the EA_RAG framework are detailed in the Jupyter Notebook ea-rag/main.ipynb. Follow the instructions in the notebook to:

• Embed knowledge graph entities using LLM-based text embedding models.
• Store and retrieve embeddings with Pinecone.
• Query and align entities using a combination of retrieval mechanisms and generative models.

EA_RAG Results

Experiments were conducted on subsets of the DBP15K dataset (English-French), comparing the RAG-based alignment against ground truth. Below are the results:

Dataset Size	Hits@1	Processing Time
150 Entities	0.6667	~30 seconds
1,500 Entities	0.6421	~7 minutes
15,000 Entities	0.6465	~61 minutes

EAkit Results

The embedding-based methods implemented in EAkit were evaluated on the same DBP15K (English-French) dataset, focusing on Hits@1 accuracy. Below are the results for each method:

Method	Hits@1	Processing Time
BootEA	0.5148	~12 hours
GCN-Align	0.4303	~25 minutes
MTransE	0.4073	~11 minutes

Running EAkit

To benchmark EA_RAG, we evaluated it against embedding-based methods provided by EAkit. Note that EAkit requires a separate environment setup:

Environment for EAkit

1. Create a Conda environment for EAkit:

   conda create --name eakit python=3.7 -y
   conda activate eakit

2. Install dependencies:

   pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
   pip install torch-geometric torch-scatter torch-sparse torch-cluster torch-spline-conv -f https://data.pyg.org/whl/torch-2.0.0+cpu.html
   pip install numpy scipy scikit-learn tensorboard

Running EAkit Methods

1. Activate the EAkit environment:

   conda activate eakit

2. Navigate to the examples folder:

   cd EAkit/examples

3. Run a predefined alignment script (e.g., GCN-Align):

   ./run_GCN-Align.sh

4. Alternatively, execute directly with custom parameters:

   python ../run.py --data_dir "../data/DBP15K/zh_en" --encoder "GCN-Align"

5. Monitor metrics using TensorBoard:

   ./Tensorboard.sh

   Open http://localhost:6006 in a browser.

---

License

This project is licensed under the MIT License. See the LICENSE file for details.

---

Citation

If you use this framework, please cite the relevant works:

 @inproceedings{bara2025retrieval,
   author    = {Davide Mario Ricardo Bara and
                Daria Maria Meseșan and
                Gheorghe Cosmin Silaghi},
   title     = {Retrieval-Augmented Generation for Entity Alignment in Knowledge Graphs: an Incipient Experiment},
   booktitle = {Advanced Information Systems Engineering Workshops, CAiSE 2025 International Workshops},
   year      = {2025},
   publisher = {Springer},
   series    = {Lecture Notes in Computer Science},
   note      = {To appear}
 }