Knowledge Graph Embedding Based on Semantic Hierarchical Spatial Rotation

The knowledge graph is a structured representation of real-world triples. Knowledge graph embedding is an effective method to predict the missing part of the knowledge graph. To better improve the knowledge graph, this paper uses a rotation matrix to embed entities and relationships into three-dimensional space and proposes a new knowledge graph embedding model-semantics Layered spatial rotation model (SROTATE). SROTATE includes an entity-based semantic layering strategy and a three-dimensional space rotation strategy. Specifically, the entity-based semantic layering strategy can layer different entities according to semantics, so that entities with different semantics are distributed on different levels. In addition, the spatial rotation strategy associates the head entity and the tail entity with the rotation matrix, thereby distinguishing entities of the same level. According to the experimental results, SROTATE has a certain performance improvement compared with mainstream algorithms in large-scale data sets and complex relationships.