关键词: 知识图谱, 关系推理, 表示学习, 深度神经网络

Abstract: Because of its powerful semantic processing function and fast analysis ability, knowledge graph has been widely used in search, question and answer, diagnosis, etc. in recent years. However, the existing technology can't automatically build a complete knowledge graph, and the relational reasoning technology is needed to fill in the missing relationships of knowledge graph. Based on the spacecraft fault knowledge graph, this paper proposes a relational reasoning method which combines representation learning and deep neural network, and takes the semantic information and position information of entities and relationships into the calculation range. Firstly, the representation learning model is used to process triples to obtain semantic vectors. Secondly, the principal component analysis method is used to reduce the dimension of semantic vector in order to reduce the difficulty of subsequent calculation. Then, according to the global position of the entity in the knowledge graph, the position vector of the entity is obtained via Boolean data markers. Finally, the semantic vector and position vector are spliced as the input of the deep neural network, and the relational prediction vector is output. This method effectively solves the problems of complex physical relations, scarce samples and incomplete knowledge base of spacecraft fault knowledge graph. Experimental results show that the prediction accuracy of this method is higher than that of single representation learning reasoning and path reasoning, and it can effectively improve the relationship of spacecraft fault knowledge graph.

Key words: knowledge graph, relational reasoning, representation learning, deep neural network