Representation Learning Of Knowledge Graph Method With Context Information

Knowledge graph is an important information database nowadays.In the future,it will continue to play a critical role in information retrieval and reasoning in retail,supply chain,finance and other fields.Although knowledge graphs have structured the representation of knowledge information,how to use this information in practical production applications is still a big challenge.Representation learning of knowledge graphs is a solution to use inner information in knowledge graphs.Representation learning of knowledge graphs has gradually become a key part of the application of knowledge graph to the production field,and can be applied to many fields such as recommendation systems,question answering,natural language understanding and so on.Most of the researchs on representation learning of knowledge graphs focuses on the information in the internal triples.Recent years,many studies have shown that the contextual information between entities in the knowledge graphs,such as hidden connections between entities,neighborhood information,path information,etc.,can play a significant role in the knowledge graph representation learning task.This thesis focuses on the research and experiments on the hidden connections and neighborhood information between entities,and explores their impact on representation learning of knowledge graphs.For the hidden connections between entities,we propose a knowledge graph representation learning translation model combined with an attention mechanism.Based on Trans D with non-linear mappings,we use attention mechanism to link some entities with others to further enhance the effectiveness of the model.In addition,the complexity of our model is lower than most neural network models,but it still has a good performance.For neighborhood information,we propose a representation learning neural network model combined with neighborhood information.Based on a neural network model,we add the neighborhood information channel.We use random walk technology to obtain the neighborhood node sequences of entities,and analyze the feature information through convolution neural network.Then we combine the neighborhood information channel with the triple information channel to generate the final prediction result.Experiments show the role of neighborhood information in knowledge graph representation learning.