KAG: Boosting LLMs in Professional Domains via Knowledge Augmented Generation

Abstract

The recently developed Retrieval-Augmented Generation (RAG) technology has enabled the efficient construction of domain-specific applications. The key technologies of RAG are retrieval based on similarity and reasoning based on next-token prediction. However, this approach differs significantly from how humans solve problems. Humans typically follow certain analytical logic, reasoning while retrieving relevant information, and then connecting the clues to serve as references, ultimately generating an answer. In this process, the focus is on the semantic type and clear relationships between the keywords rather than similarity and co-occurrence. This difference in methodology results in the answers generated by RAG technology being insufficiently accurate or valuable.