Can knowledge graphs enable multi-hop reasoning in one retrieval step?
Standard RAG retrieves once but misses chains; iterative RAG follows chains but costs more. Can we encode multi-hop paths in a knowledge graph so one retrieval pass discovers them all?
Multi-hop question answering requires finding information that depends on other information: "Who was the mentor of the person who founded X?" requires finding X's founder, then finding that person's mentor. Standard RAG retrieves once per question. Iterative RAG retrieves multiple times, issuing new queries based on intermediate answers. Both approaches have costs: single-step RAG misses the chain; iterative RAG is slow and expensive.
HippoRAG takes a third approach, inspired by hippocampal indexing theory of human memory. The corpus is converted into a schemaless knowledge graph: LLMs extract entities and relationships from passages, forming a graph where nodes are concepts and edges are relationships. This happens offline, at indexing time.
At query time: extract key concepts from the query; seed Personalized PageRank (PPR) on the knowledge graph using those concepts. PPR propagates importance scores through the graph from the seed nodes, naturally following edges (relationships) and discovering connected subgraphs. Passages associated with highly-ranked graph regions are retrieved.
The multi-hop capability emerges naturally from graph traversal: PPR follows relationship edges, which means it discovers concepts multiple hops away from the query seeds. The traversal is complete in a single retrieval step because the graph encodes all multi-hop paths.
Results: 20% better than state-of-the-art on multi-hop QA; single-step HippoRAG matches or beats iterative retrieval (IRCoT) while being 10-20x cheaper and 6-13x faster. Integrating HippoRAG into IRCoT brings further gains.
The memory analogy holds architecturally: the knowledge graph functions as an associative index (analogous to hippocampus) over perceptual content (analogous to neocortex), enabling pattern-completion retrieval from partial query cues.
Complementary KG-based approaches: GraphRAG (Microsoft) uses knowledge graphs for a fundamentally different purpose than HippoRAG. Where HippoRAG uses the KG for traversal-based retrieval (find specific facts via graph navigation), GraphRAG uses KG + Leiden community detection for modular summarization — answering global corpus questions ("What are the main themes?") that no single retrieval step can address. The distinction: HippoRAG exploits graph connectivity; GraphRAG exploits graph modularity. See Can community detection enable RAG systems to answer global corpus questions?. SymAgent takes yet a third approach: deriving symbolic rules from KG structure to create navigational plans that align natural language with graph topology — using the graph not for retrieval or summarization but for structural reasoning guidance. See Can symbolic rules from knowledge graphs guide complex reasoning?.
Inquiring lines that use this note as a source 25
This note is a source for these synthesized inquiries. Follow a line forward into its question, or open it to trace back to all of its sources.
- What graph structures better support multi-hop reasoning than pairwise edges?
- How does structure-aware retrieval routing differ from existing graph-versus-vector RAG tradeoffs?
- Can knowledge graph structure help embeddings represent more combinations?
- Why does community detection in knowledge graphs outperform pure retrieval or pure summarization?
- How should enterprises choose between graph and vector approaches for RAG?
- How do community-based summaries differ from retrieval-based traversal in knowledge graph RAG?
- How does map-reduce over communities compare to flat multi-hop retrieval architectures?
- How does query planning as a separate step improve multi-hop retrieval coherence?
- Can inference-time query decomposition replace pre-built knowledge graph structures?
- How does hypergraph accumulation differ from single-pass graph retrieval?
- How do hierarchical query planning architectures improve multi-hop retrieval?
- How does knowledge graph structure enable multi-hop reasoning in recommendations?
- How does GraphRAG differ from HippoRAG despite both using knowledge graphs?
- Can query-time logic graphs match the efficiency of pre-built knowledge graph indexing?
- When should you use knowledge graphs instead of semantic vector retrieval systems?
- Why are pairwise relations insufficient for representing higher-order multi-hop reasoning?
- Can graph-based retrieval with knowledge graphs scale to multi-hop reasoning?
- How can knowledge graphs improve over pure embedding retrieval?
- Can knowledge graph structure be exploited for efficient multi-hop retrieval?
- How do hierarchical architectures improve multi-hop query performance?
- Can single-hop knowledge automatically compose into multi-hop capability?
- How do hierarchical research architectures improve multi-hop query accuracy?
- Can sparse attention methods be designed specifically for multi-hop reasoning tasks?
- Why do aggregation tasks degrade faster than multi-hop reasoning under sparsity?
- How should retrieval systems handle multi-hop reasoning and iterative information needs?
Related concepts in this collection 4
This note in its neighbourhood — explore the map, then jump to a related concept in the list below.
Click a node to walk · click center to open · click Open in graph to see this note in the full knowledge graph
-
Can query-time graph construction replace pre-built knowledge graphs?
Does building dependency graphs from individual queries at inference time offer a more flexible and cost-effective alternative to constructing knowledge graphs over entire document collections upfront?
LogicRAG inverts this: query-time graph vs corpus-time graph; HippoRAG pre-builds; both achieve relational reasoning via graph traversal with different cost/flexibility trade-offs
-
When do graph databases outperform vector embeddings for retrieval?
Vector similarity struggles with aggregate and relational queries that require traversing multiple entity connections. Can graph-oriented databases with deterministic queries solve this failure mode in enterprise domain applications?
same architectural principle; HippoRAG implements it with PPR over a schemaless KG rather than a structured graph database
-
Can community detection enable RAG systems to answer global corpus questions?
Standard RAG struggles with corpus-wide questions that require understanding overall themes rather than retrieving specific passages. Can graph community detection overcome this limitation at scale?
complementary KG approach: community-based summarization vs traversal-based retrieval
-
Can symbolic rules from knowledge graphs guide complex reasoning?
Can deriving symbolic rules directly from knowledge graph structure help align natural language questions with structured reasoning paths? This explores whether explicit structural patterns outperform semantic similarity for multi-hop inference.
third approach: KG structure for navigational reasoning guidance
Related papers in this collection 8
Papers most semantically related to this note, ranked by cosine similarity in the embedding space.
- Weak-to-Strong GraphRAG: Aligning Weak Retrievers with Large Language Models for Graph-based Retrieval Augmented Generation
- You Don't Need Pre-built Graphs for RAG: Retrieval Augmented Generation with Adaptive Reasoning Structures
- MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries
- Large Language Models Meet Knowledge Graphs for Question Answering: Synthesis and Opportunities
- Towards Agentic RAG with Deep Reasoning: A Survey of RAG-Reasoning Systems in LLMs
- HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models
- Learning to Retrieve Reasoning Paths over Wikipedia Graph for Question Answering
- DeepDive: Advancing Deep Search Agents with Knowledge Graphs and Multi-Turn RL
Original note title
knowledge graph plus Personalized PageRank achieves multi-hop reasoning in a single retrieval step