Knowledge Map Completion Method Based on Metric Space and Relational Path

Knowledge Graph describes entities and their attributes and relationships in the objective world in a structured way. Aiming at the problem of knowledge imperfection widely existing in the knowledge map, a knowledge representation learning algorithm — PTranSparse, which is based on metric space and relational path, is proposed to complete the knowledge map. This method combines the ability of TranSparse model to process the heterogeneity and unbalance of entity and relationship and PTransE to make full use of the semantic information of relational path to improve the discrimination of knowledge representation learning. Based on the combination of the two models, the relationship types are considered, and the weights associated with the relationship types are added to differentiate the relationship types when entities are projected. Experiments show that, compared with the original models and the existing combination models and methods, this method can effectively improve the link prediction efficiency of the knowledge graph while solving the complex relation reasoning, and ensure a higher accuracy.