You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Large language models (LLMs) demonstrate powerful capabilities, but theystill face challenges in practical applications, such as hallucinations, slowknowledge updates, and lack of transparency in answers. Retrieval-AugmentedGeneration (RAG) refers to the retrieval of relevant information from externalknowledge bases before answering questions with LLMs. RAG has been demonstratedto significantly enhance answer accuracy, reduce model hallucination,particularly for knowledge-intensive tasks. By citing sources, users can verifythe accuracy of answers and increase trust in model outputs. It alsofacilitates knowledge updates and the introduction of domain-specificknowledge. RAG effectively combines the parameterized knowledge of LLMs withnon-parameterized external knowledge bases, making it one of the most importantmethods for implementing large language models. This paper outlines thedevelopment paradigms of RAG in the era of LLMs, summarizing three paradigms:Naive RAG, Advanced RAG, and Modular RAG. It then provides a summary andorganization of the three main components of RAG: retriever, generator, andaugmentation methods, along with key technologies in each component.Furthermore, it discusses how to evaluate the effectiveness of RAG models,introducing two evaluation methods for RAG, emphasizing key metrics andabilities for evaluation, and presenting the latest automatic evaluationframework. Finally, potential future research directions are introduced fromthree aspects: vertical optimization, horizontal scalability, and the technicalstack and ecosystem of RAG.
URL
Affiliations
Abstract
Translation (by gpt-3.5-turbo)
Summary (by gpt-3.5-turbo)
The text was updated successfully, but these errors were encountered: