Skip to content
Projects

waetr/KET-RAG has 205 GitHub stars and 19 forks. GitHot measured 0 stars in 24 hours. Last checked 2026-11-18 10:15 UTC. Last push 2025-06-02.

How these counts are measured

Measured star growth

in 24h
0
in 7 days
Collecting history
in 15 days
Collecting history
in 1 month
Collecting history

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

About

KET-RAG is a powerful and flexible framework for retrieval-augmented generation (RAG) enhanced with knowledge graphs. This project allows for structured document indexing and efficient LLM-based answer generation.

Developer Tools

README.md

Original

KET-RAG: Knowledge-Enhanced Text Retrieval Augmented Generation

KET-RAG is a powerful and flexible framework for retrieval-augmented generation (RAG) enhanced with knowledge graphs. This project allows for structured document indexing and efficient LLM-based answer generation.

Overview

KET-RAG balances retrieval quality and efficiency with a multi-granular indexing framework consisting of:

  • Knowledge Graph Skeleton (SkeletonRAG): Selects key text chunks via PageRank and extracts structured knowledge using LLMs.
  • Text-Keyword Bipartite Graph (KeywordRAG): Links keywords to text chunks, mimicking knowledge graph relationships with minimal cost.

During retrieval, KET-RAG integrates information from both entity and keyword channels, enabling efficient and high-quality LLM-based answer generation. Experiments show that KET-RAG significantly reduces indexing costs while improving retrieval and generation quality, making it a practical solution for large-scale RAG applications.

Prerequisites

Ensure you have Python >=3.10 installed.

Installation

Install dependencies using Poetry:

pip install poetry
poetry install

Setup

Using the folder ragtest-musique as an example, follow these steps:

1. Initialize the Project

python -m graphrag init --root ragtest-musique/

This command sets up the necessary file structure and configurations.

2. Tune Prompts

python -m graphrag prompt-tune --root ragtest-musique/ --config ragtest-musique/settings.yaml --discover-entity-types

Adjust prompts for better retrieval.

3. Build the Index

Before running this step, modify settings.yaml to set the appropriate parameters as needed, based on our paper.

python -m graphrag index --root ragtest-musique/

This process creates an indexed structure for retrieval.

Generate Context and Answers

Environment Setup

Before executing the scripts, set up your API key:

export GRAPHRAG_API_KEY=your_api_key_here

1. Generate Context

To generate a single context file:

python indexing_sket/create_context.py ragtest-musique/ keyword 0.5

Parameters:

  • First argument: Root directory of the project
  • Second argument: Context-building strategy (text, keyword, or skeleton)
  • Third argument: Context threshold theta (range: 0.0-1.0)

2. Generate Answers

To generate answers for all context files in the output directory:

python indexing_sket/llm_answer.py ragtest-musique/

Acknowledgments

This project builds upon Microsoft's GraphRAG (version 0.4.1), licensed under the MIT License.