HippoRAG

HippoRAG is a neurobiologically-inspired RAG framework that uses knowledge graphs and Personalized PageRank (PPR) for single-step, multi-hop retrieval, delivering up to 20% better performance on QA benchmarks.

HippoRAG is a novel Retrieval-Augmented Generation (RAG) framework, directly inspired by the hippocampal indexing theory of human long-term memory. The system synergistically orchestrates Large Language Models (LLMs), knowledge graphs (KG), and the Personalized PageRank (PPR) algorithm to enable efficient, single-step multi-hop retrieval (a key RAG challenge). Specifically, the LLM builds a KG—the 'artificial hippocampal index'—and PPR traverses it to integrate information across documents. This neurobiological approach yields significant performance gains: it is 6–13 times faster and 10–30 times cheaper than iterative retrieval methods like IRCoT, while boosting accuracy on multi-hop QA benchmarks by up to 20%.

https://github.com/OSU-NLP-Group/HippoRAG