HippoRAG

Primary use cases

• multi-hop (associative) question answering over a document corpus
• knowledge-graph-augmented retrieval pipelines
• sense-making over long / interconnected contexts
• research baselines comparing graph retrieval to dense top-k and to other graph-RAG systems (GraphRAG, RAPTOR, LightRAG)

Key concepts

• Hippocampal indexing theory → hippocampus-rag (docs) — Cognitive-neuroscience theory that the hippocampus stores sparse pointers (indices) into neocortical patterns, used as the design metaphor for HippoRAG's KG + retrieval split.
• OpenIE knowledge graph → graphrag (docs) — Documents are decomposed by an LLM into entity-relation triples that are merged into a persistent knowledge graph with vector embeddings on nodes (the offline indexing phase).
• Personalized PageRank retrieval → hippocampus-rag (docs) — At query time, entities extracted from the question seed Personalized PageRank over the KG; top-scoring nodes' associated passages are returned to the reader LLM.
• Neocortex / hippocampus analogy → hippocampus-rag (docs) — The framework's design metaphor: the LLM plays the neocortex (pattern completion / language) and the KG + PPR plays the hippocampus (associative pointer index).
• Associative / factual / sense-making memory (docs) — HippoRAG 2 explicitly targets three memory regimes: associative (multi-hop), factual, and sense-making (integrating large complex contexts), reporting gains over baseline RAG on each.
• HippoRAG Python API (docs) — Library surface: HippoRAG(save_dir, llm_model_name, embedding_model_name).index(docs), .retrieve(queries), .rag_qa(queries) — combined or separate retrieval-and-QA calls.