JECI++: A Modified Joint Knowledge Graph Embedding Model for Concepts and Instances

Abstract

Abstract Concepts and instances are important parts in knowledge graphs, but most knowledge graph embedding models treat them as entities equally, that leads to inaccurate embeddings of concepts and instances. Aiming to solve this problem, we propose a novel knowledge graph embedding model called JECI++ to jointly embed concepts and instances. First, JECI++ simplifies hierarchical concepts based on subClassOf relation and instanceOf relation, then re-links instances to the simplified concepts as new instanceOf triples. Consequently, an instance can be obtained by its neighbor instances and its belonging simplified concepts. Second, circular convolution is utilized to locate an instance in the embedding space, based on neighbor instances and simplified concepts. Finally, simplified concepts and instances are jointly embedded by the embeddings learner with CBOW (Continuous Bag-of-Words) or Skip-Gram strategies. Especially, JECI++ can alleviate the problem of complex relations by incorporating neighbor information of instances. JECI++ is evaluated by link prediction and triple classification on real world datasets. Experimental results demonstrate that it outperforms state-of-the-art models in most cases.