Multi-Hop Reasoning

Multi-hop reasoning is the ability to answer questions that require following chains of connected information across multiple intermediate steps. For example, answering "What technology does Alice's manager's company use?" requires three hops: Alice -> manager (Bob) -> company (Acme Corp) -> technology (Python, AWS).

Multi-hop reasoning is a primary motivation for combining knowledge graphs with RAG systems. Pure RAG struggles with multi-hop questions because each retrieval step is independent -- it cannot reliably chain information across separately retrieved documents. Knowledge graphs excel at multi-hop reasoning because the chain of relationships is explicitly represented and can be traversed in a single query: MATCH (a:Person {name:'Alice'})-[:REPORTS_TO]->(m)-[:WORKS_FOR]->(c)-[:USES]->(t) RETURN t.

In hybrid RAG+KG systems, multi-hop reasoning typically starts with knowledge graph traversal to establish the chain of entities and relationships, then uses RAG retrieval to find supporting details and context for each entity in the chain. This combination produces answers that are both structurally correct (from the graph) and richly detailed (from the documents).