Multi-Hop Retrieval
Definition
Multi-Hop Retrieval is a retrieval process in which answering an information need requires multiple retrieval steps, with information from one step informing what must be retrieved next.
Why it matters
Some questions cannot be answered from one document or one retrieval operation. Multi-hop retrieval allows a system to connect facts, entities or evidence across multiple sources or stages.
How it works
Initial retrieval
- The system retrieves information relevant to the first part of the information need.
Intermediate reasoning or extraction
- Information from the first retrieval step identifies another entity, relationship, constraint or missing fact.
- That intermediate result informs the next retrieval request.
Subsequent retrieval
- Additional searches or retrieval steps gather the remaining evidence.
- The process can continue until the system has enough information or reaches a stopping condition.
How Netsleek uses the term
Netsleek uses Multi-Hop Retrieval as the established retrieval concept for chained evidence gathering. It is kept distinct from Query Fan-Out, which can run multiple queries in parallel, and from Agentic Retrieval, which is the broader agent-driven retrieval behaviour.
Comparisons
- Multi-Hop Retrieval vs Query Fan-Out: Fan-out generates multiple retrieval paths from an originating need. Multi-hop retrieval uses information from one retrieval step to determine a later step.
- Multi-Hop Retrieval vs Multi-Query Decomposition: Decomposition splits a complex problem into sub-queries. Multi-hop retrieval specifically describes sequential dependency between retrieval steps.
- Multi-Hop Retrieval vs Agentic Retrieval: Agentic Retrieval can include multi-hop behaviour, but also includes tool selection, iteration and other control logic.
How Multi-Hop Retrieval Connects Across AI Systems
Multi-Hop Retrieval does not operate in isolation. It connects to other established concepts across the AI search and retrieval process:
- Can use entity relationships: Entity Linking and Knowledge Graph relationships can reveal the next retrieval target.
- Can use tools: AI Agents can execute sequential search or retrieval steps.
- Supports complex RAG: RAG systems can retrieve evidence across several dependent steps.
- Builds evidence chains: Multiple retrieved sources can contribute to one final answer or claim.
Related glossary concepts
- Agentic Retrieval
- Query Fan-Out
- Multi-Query Decomposition
- Entity Linking
- Knowledge Graph
- Retrieval-Augmented Generation (RAG)
- Tool Use (Function Calling)
Common misinterpretations
- Multi-hop retrieval is not simply retrieving several documents at once.
- Multiple queries do not automatically constitute multi-hop retrieval.
- Sequential retrieval does not guarantee that the final reasoning chain is correct.
Summary
Multi-Hop Retrieval gathers information through multiple dependent retrieval steps. It is useful for complex questions that require connecting facts or entities across more than one source or retrieval operation.