Rot-Pro: Modeling Transitivity by Projection in Knowledge Graph Embedding

Knowledge graph embedding models learn the representations of entities and\nrelations in the knowledge graphs for predicting missing links (relations)\nbetween entities. Their effectiveness are deeply affected by the ability of\nmodeling and inferring different relation patterns such as symmetry, asymmetry,\ninversion, composition and transitivity. Although existing models are already\nable to model many of these relations patterns, transitivity, a very common\nrelation pattern, is still not been fully supported. In this paper, we first\ntheoretically show that the transitive relations can be modeled with\nprojections. We then propose the Rot-Pro model which combines the projection\nand relational rotation together. We prove that Rot-Pro can infer all the above\nrelation patterns. Experimental results show that the proposed Rot-Pro model\neffectively learns the transitivity pattern and achieves the state-of-the-art\nresults on the link prediction task in the datasets containing transitive\nrelations.\n