Research On Knowledge Graph Completion By Combining Structural And Semantic Information

As knowledge graph provides structured sematic information that can be understood by computers, it's an indispensable part of building AI systems.Although some existing large-scale knowledge graphs stored a lot of facts, they are far from complete. So the technology of automatic knowledge graph completion is of great significance.This paper first compares the popular embedding models and graph feature models in the task of knowledge graph completion, and proposes that more accurate embeddings can be learned by merging the structure and semantic information of the knowledge graph. And then proposes two new models by introducing the structural features into the classical knowledge graph embedding model from two different perspectives. One is translating embedding model with triangle motifs, and the other is translating embedding model with structure based semantic similarity. The experimental results show that the combination of structural and semantic information is conducive to the accuracy of knowledge graph embedding.In addition, this paper also explores a new application of the knowledge graph by designing and building a question answering system for tourist sights recommendation based on the specific domain knowledge graph.