Prior Bilinear-Based Models for Knowledge Graph Completion

Bilinear-based models are powerful and widely used approaches for Knowledge Graphs Completion (KGC). Despite the considerable progress achieved by bilinear-based models, prior research has predominantly focused on posterior properties, such as symmetry patterns, while neglecting the consideration of prior properties. In this paper, we identify a prior property known as "the law of identity" that eludes capture by bilinear-based models, thus impeding their comprehensive modeling of Knowledge Graph (KG) characteristics. To overcome this limitation, we propose a novel solution named Unit Ball Bilinear Model (UniBi). UniBi not only attains theoretical superiority but also enhances interpretability and performance by minimizing ineffective learning through minimal constraints. Experimental results demonstrate that UniBi effectively models the prior property while validating its interpretability and performance.