# Corrective Retrieval Augmented Generation

**Authors:** Shi-Qi Yan, Jia-Chen Gu, Yun Zhu, Zhen-Hua Ling  
**arXiv:** [2401.15884](https://arxiv.org/abs/2401.15884)  
**Date:** 2024-01-29  
**Summary page:** https://www.summarizepaper.com/en/paper/2401.15884v1/

## Key points

- The paper introduces the concept of Corrective Retrieval Augmented Generation (CRAG) to improve text generation by large language models (LLMs).
- CRAG addresses limitations of retrieval-augmented generation (RAG) by incorporating a lightweight retrieval evaluator.
- CRAG leverages large-scale web searches to augment results from static and limited corpora.
- CRAG utilizes a decompose-then-recompose algorithm to filter out irrelevant information and focus on key information.
- The proposed approach can be easily integrated with various RAG-based approaches.
- Experimental results on four datasets show that CRAG significantly improves performance for both short- and long-form generation tasks.
- CRAG presents an innovative solution to enhance text generation by addressing challenges faced by LLMs.

## Summary

The paper introduces the concept of Corrective Retrieval Augmented Generation (CRAG) as a solution to improve the accuracy and reliability of text generation by large language models (LLMs). CRAG addresses the limitations of retrieval-augmented generation (RAG) by incorporating a lightweight retrieval evaluator that assesses the quality of retrieved documents and triggers different knowledge retrieval actions. It also leverages large-scale web searches to augment the results from static and limited corpora. Additionally, CRAG utilizes a decompose-then-recompose algorithm to filter out irrelevant information and focus on key information in retrieved documents. The proposed approach can be easily integrated with various RAG-based approaches, making it a plug-and-play solution. Experimental results on four datasets show that CRAG significantly improves the performance of RAG-based approaches for both short- and long-form generation tasks. In summary, this paper presents an innovative solution to enhance text generation by addressing the challenges faced by LLMs through corrective retrieval augmented generation.