A Representation Learning Method of Fusing Entity Affinity Constraints

Representation learning on knowledge graph aims to project both entities and relations into a low-dimensional continuous space and dig out the hidden relations between two entities. Traditional method does not make full use of entity's description text and most of representation learning methods based on entity description project text into vector space without considering the relevance of entities in text. In this paper, a knowledge graph representation learning method was proposed, taking the advantage of entity description based on the traditional structure-based representation learning. In this method, the different relevant entities extracted based on entities description and relevant entities were fused as supplementary constraints information to knowledge graph representation learning. Experimental results on real world datasets show that, this method can enhance the inference effectiveness and outperforms structure-based representation learning method, especially outperform deep convolutional neural model which encode semantics of entity descriptions into structure-based representation learning.