Knowledge Graph Inference using Tensor Embedding

Axiom based inference provides a clear and consistent way of reasoning to add more information to a knowledge graph. However, constructing a set of axioms is expensive and requires domain expertise, time, and money. It is also difficult to reuse or adapt a set of axioms to a knowledge graph in a new domain or even in the same domain but using a slightly different representation approach. This work makes three main contributions, it (1) provides a family of representation learning algorithms and an extensive analysis on eight datasets; (2) yields better results than existing tensor and neural models; and (3) includes a provably convergent factorization algorithm.