How do graph databases handle traversals differently from relational joins?

Graph databases (Neo4j, Neptune) store relationships as first-class citizens with direct pointers between nodes - traversals follow pointers with O(1) per hop regardless of data size. Relational joins compute relationships at query time using indexes - cost grows with data size. Graph advantages: variable-length paths, pattern matching (Cypher, Gremlin), relationship properties. Index-free adjacency enables efficient multi-hop traversals. Best for highly connected data: social networks, recommendations, fraud detection.