Knowledge Representation Learning Based On SoftMax

Knowledge Graph (KG) embedding has begun to be widely used in various fields, such as search engine, semantic search, and knowledge inference. In the traditional model, because an effective time period of knowledge is not taken into account, it leads to a wrong knowledge construction. And their robustness is not excellence after adding new knowledge elements. In most existing KG representation learning models, repeated calculations are performed by synthesizing a negative sample set with the correct sample set, which also leads to an increase in the amount of calculation. At the same time, because there are many-to-many situations for entities and relationships in the KG, the vector representation obtained when using the previous model for representation learning is not unique. So it leads to an increase in computational complexity. Therefore, in this paper we propose a knowledge representation learning model based on softmax, which calculates the most suitable representation vector by the conditional probability of the entity and relation vector. The experimental results show that our model has improved significantly in the knowledge completion compared with the traditional method.