# HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models

**Authors:** Bernal Jiménez Gutiérrez, Yiheng Shu, Yu Gu, Michihiro Yasunaga, Yu Su  
**arXiv:** [2405.14831](https://arxiv.org/abs/2405.14831)  
**Date:** 2024-05-23  
**Summary page:** https://www.summarizepaper.com/en/paper/2405.14831v2/

## Key points

- Paper titled "HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models"
- Introduces HippoRAG retrieval framework based on hippocampal indexing theory
- Synergistically combines LLMs, knowledge graphs, and Personalized PageRank algorithm
- Outperformed existing RAG methods by up to 20% in multi-hop question answering tasks
- Single-step retrieval with HippoRAG achieved comparable or superior performance to iterative approaches like IRCoT
- Integration of HippoRAG into IRCoT led to substantial performance gains
- Offers solution for enhancing knowledge integration in large language models
- Code and data available on GitHub at https://github.com/OSU-NLP-Group/HippoRAG

## Summary

, , , , 

In their paper titled "HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models," authors Bernal Jiménez Gutiérrez, Yiheng Shu, Yu Gu, Michihiro Yasunaga, and Yu Su delve into the challenges faced by large language models (LLMs) in efficiently integrating new experiences after pre-training. They introduce HippoRAG, a novel retrieval framework based on the hippocampal indexing theory of human long-term memory that aims to enable deeper and more efficient knowledge integration over new experiences. This framework synergistically combines LLMs, knowledge graphs, and the Personalized PageRank algorithm to mimic the roles of the neocortex and hippocampus in human memory. The authors conducted experiments comparing HippoRAG with existing retrieval-augmented generation (RAG) methods on multi-hop question answering tasks. The results showed that HippoRAG outperformed state-of-the-art methods by up to 20%. Notably, single-step retrieval with HippoRAG achieved comparable or superior performance compared to iterative retrieval approaches like IRCoT while being significantly more cost-effective and faster. Furthermore, integrating HippoRAG into IRCoT led to substantial performance gains. The authors demonstrated that their method could tackle new types of scenarios that were previously out of reach for existing methods. Overall, HippoRAG offers a promising solution for enhancing knowledge integration in large language models and represents a significant advancement in the field of natural language processing. The code and data for implementing HippoRAG are available on GitHub at https://github.com/OSU-NLP-Group/HippoRAG.