From: How Entities and Relations Become Vectors

When evaluating an embedding model, which relation patterns does your dataset actually contain?

Model choice should follow graph structure, not popularity. A translational model may behave differently on symmetric, hierarchical, or one-to-many relations.

Action

Audit relation cardinalities and symmetry before selecting an architecture.

Read the full exploration

What else is in this exploration

3 evidence blocks3 perspectives2 visualizations8 media resources4 rabbit holes

evidence

TransE represents a relation as a translation between the vectors of two entities.

evidence

Knowledge graph embeddings map entities and relations into continuous vector spaces.

evidence

Embedding training learns vectors by contrasting true triples with corrupted triples.

Continue exploring

How Entities and Relations Become Vectors

Evidence, perspectives, rabbit holes, and more