Relationship Label

A Relationship Label is the name assigned to an edge (connection) in a knowledge graph that describes the type of relationship between two nodes. Relationship labels determine HOW to link, not just WHAT to link. Examples include HAS_ATTRIBUTE (entity has an attribute), SHARES_ATTRIBUTE (two entities share the same attribute), or IS_PART_OF (hierarchical relationship). For instance, SHARES_ATTRIBUTE suggests a comparison article, HAS_ATTRIBUTE suggests a section in the main article, IS_PART_OF suggests hierarchical linking (parent → child). Labeled relationships are the fundamental advantage of graphs over simple cosine similarity: embeddings show similarity, but graphs explain the reason—they share attribute X.

For example, the relationship SHARES_ATTRIBUTE('side_effects') between 'cortisol' and 'prednisone' justifies a link with anchor text 'corticosteroid side effects'.

Consistent labeling across the entire graph is essential for automated analysis. Chaotic naming (HAS vs OWNS vs CONTAINS) makes automated analysis extremely difficult.