URAG: Unified Retrieval-Augmented Generation
Yulun Song, Long Yan, Lina Qin, Gongju Wang, Xingru Huang, Luzhe Hu, Weixin Liu · 2024
To address the issues of insufficient retrieval capabilities and "hallucinations" in responses generated by large models, this paper proposes a knowledge question-answering framework based on Unified Retrieval-Augmented Generation (URAG).The framework integrates three retrieval mechanisms-keyword retrieval, vector retrieval, and graph retrieval-enabling efficient and high-quality utilization of massive, multi-source, and heterogeneous data.It effectively overcomes the limitations of a single retrieval pathway.Experimental results demonstrate that the URAG framework excels across various task scenarios, enhancing the accuracy and comprehensiveness of knowledge retrieval.Its advantage is particularly evident when dealing with multi-dimensional and multi-layered information.In conclusion, the Unified Retrieval-Augmented Generation technique offers a new method for improving the performance of intelligent question-answering systems, with broad application prospects and research value.This study provides valuable insights into understanding RAG (Retrieval-Augmented Generation) technology and its application in large language models.