The paper discusses TWIG (Topologically-Weighted Intelligence Generation), an innovative paradigm for simulating the output of Knowledge Graph Embeddings (KGEs) using a significantly lesser number of parameters. TWIG learns from inputs comprising topological features of graph data, without coding for latent representations of entities or edges. The authors claim that hyperparameter choice in KGEs is a deterministic function of the KGE model and graph structure, and that KGEs do not learn latent semantics, but only latent representations of structural patterns. The study suggests that TWIG can be used to accurately predict new facts in KGs without needing node and edge embeddings.
Publication date: 12 Feb 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2402.06097
