# WeKnow-RAG: An Adaptive Approach for Retrieval-Augmented Generation Integrating Web Search and Knowledge Graphs

**Authors:** Weijian Xie, Xuefeng Liang, Yuhui Liu, Kaihua Ni, Hong Cheng, Zetian Hu  
**arXiv:** [2408.07611](https://arxiv.org/abs/2408.07611)  
**Date:** 2024-08-14  
**Summary page:** https://www.summarizepaper.com/en/paper/2408.07611v1/

## Key points

- Large Language Models (LLMs) are a significant advancement in artificial intelligence for developing adaptive intelligent agents.
- The ultimate goal is to achieve Artificial General Intelligence (AGI.
- Challenges faced by LLMs include generating factually incorrect information and "phantom" content, affecting their reliability and usability.
- WeKnow-RAG is a novel approach that integrates Web search capabilities and Knowledge Graphs into a Retrieval-Augmented Generation (RAG) framework.
- WeKnow-RAG aims to enhance the accuracy and reliability of LLM responses by combining structured representation with dense vector retrieval, leveraging domain-specific knowledge graphs for a wide range of queries and domains.
- Multi-stage web page retrieval techniques are utilized in WeKnow-RAG to improve performance on tasks requiring factual information and complex reasoning through both sparse and dense retrieval methods.
- WeKnow-RAG strikes a balance between efficiency and accuracy in information retrieval, enhancing the overall process.
- The system incorporates a self-assessment mechanism for LLMs to evaluate the trustworthiness of generated answers, further enhancing reliability for users.
- WeKnow-RAG has been demonstrated through offline experiments and online submissions to deliver accurate and reliable information across various domains, showcasing its potential as a valuable tool in advancing AI research and applications.

## Summary

In the realm of artificial intelligence, Large Language Models (LLMs) have emerged as a significant advancement in the development of adaptive intelligent agents. The ultimate goal is to achieve Artificial General Intelligence (AGI). However, despite their potential, LLMs face challenges such as generating factually incorrect information and "phantom" content. This undermines their reliability and usability in real-world scenarios. To address these issues, a novel approach called WeKnow-RAG has been proposed. WeKnow-RAG integrates Web search capabilities and Knowledge Graphs into a Retrieval-Augmented Generation (RAG) framework. By combining the structured representation of Knowledge Graphs with the flexibility of dense vector retrieval, it aims to enhance the accuracy and reliability of LLM responses. Additionally, WeKnow-RAG leverages domain-specific knowledge graphs to cater to a wide range of queries and domains. This approach improves performance on tasks requiring factual information and complex reasoning by utilizing multi-stage web page retrieval techniques that utilize both sparse and dense retrieval methods. By striking a balance between efficiency and accuracy in information retrieval, WeKnow-RAG enhances the overall process. Furthermore, WeKnow-RAG incorporates a self-assessment mechanism for LLMs to evaluate the trustworthiness of their generated answers. This feature further enhances the system's reliability and ensures that users can have confidence in the provided information. The effectiveness of WeKnow-RAG has been demonstrated through offline experiments and online submissions. The system has proven its capability to deliver accurate and reliable information across various domains, showcasing its potential as a valuable tool in advancing AI research and applications.