Hongjin Qian
Researcher, Wisdom Source Institute
He is responsible for the research work in the field of large language model retrieval enhancement.He graduated from the School of Artificial Intelligence, Renmin University of China (RUC) in 2024 with a PhD degree, and his research interests are in natural language processing and information retrieval.He has published several papers in international conferences such as ACL, EMNLP, SIGIR, theWebConf, etc.He was nominated for the best paper in theWebConf 2023.
Topic
RAG 2.0: Memory-Driven Next Generation Retrieval Enhancement System
Conventional RAG (Retrieval Augmented Generation) systems lack the ability to globally memorise retrieved databases, limiting their application mainly to direct question and answer tasks. We draw on human memory mechanisms to explore memory models capable of global semantic awareness of very long texts, helping to achieve a quick overview and mastery of the complete context. In complex tasks requiring global information comprehension, memory-driven RAG systems can provide more comprehensive and accurate information feedback, thus significantly improving task performance and generation quality. Outline: 1.Introduction to RAG 2.Introduction to the classical RAG system based on Embedding retrieval method 3.Introduction to RAG2.0 system based on memory mechanism 4.Application scenarios and implementation of RAG2.0 based on memory mechanism
