Research On Prediction And Recommendation Of Knowledge Graph Relationship Based On Deep Learning

Research on the knowledge graph relationship prediction method based on deep learning can produce greater practical application value for the reasoning of the relationship between nodes and the mining of hidden information.It can be used in crime prediction,recommendation system,natural language reasoning and other systems or methods.This paper studies the relationship prediction method based on the knowledge graph and applies it to the recommendation system to improve the accuracy.This paper mainly solves the problem of inaccurate recommendation caused by insufficient knowledge graph information mining.First of all,in order to better utilize and mine the graph information and get a more accurate expression of the graph relationship,this paper proposes a relationship prediction method based on RNN network and Rotat E strategy.Secondly,in order to solve the problem of inaccurate recommendation caused by insufficient expression of knowledge graph information,this paper proposes a recommendation method based on the expression of polarization relations.Finally,in order to capture the global information in the graph and further improve the recommendation accuracy,this paper proposes an adaptive reward sampling method based on reinforcement learning.Related works is as follows:1?For the relationship prediction methods based on knowledge graphs,path-oriented strategies are usually used and the differences in node neighborhood relationships are ignored,that will leads to excessive path redundancy information,which reduces the prediction accuracy.This paper proposes a method based on RNN network and Rotat E strategy,which improves the accuracy of relationship prediction by embedding learning and accurate expression of node neighborhood relationships.First,the RNN network is used to effectively learn the data association characteristics of different node neighborhoods,so that the parameters include neighborhood node information.Then,the Rotat E score strategy is used to describe the difference of node relationship,so that the node relationship of the encoding end is more clearly distinguished.In order to make full use of the above methods and solve the problem of insufficient representation of graph information in traditional recommendation algorithms,this paper has done the following work.2?In the recommendation system,the traditional method is not accurate enough to express the relationship between nodes,and at the same time,the problem of the relationship characteristics hidden by the low-dimensional data between nodes is often ignored.In order to improve the accuracy of recommendation,this paper proposes a new expression method based on the polarization relationship,which uses knowledge graph expression between nodes to the unitary space.This method enriches the effective information of relationship expression between nodes.In addition,a method for associative learning of low-dimensional data in the knowledge graph embedding and recommendation process is designed,and the hidden rich and detailed relationships are deeply explored,thereby improving the accuracy of recommendation.In order to further improve the accuracy of recommendation and solve the problem of insufficient mining of the global information of the graph in the recommendation process,this paper has done the following work.3?In the recommendation system,the traditional sampling based on reinforcement learning is not accurate enough.This paper proposes a new adaptive weight distribution reward function that integrates the positive and negative sample states of sampling points and purchase behavior,and solves the problem of inaccurate target transfer caused by uneven weight distribution.At the same time,this paper reduces the dependence of sampling points on neighbor nodes by integrating the strategy network of graph aggregation and graph convolution,solves the problem of insufficient learning of structure and attribute information in the graph network,and improves the accuracy of recommendation.In the above work,the knowledge graph representation method based on polarization relationship proposed in this paper is the most important contribution.This method solves the problem of inaccurate relationship prediction and recommendation caused by insufficient expression of map information.At the same time,in the application research of knowledge graph relational reasoning based on deep learning,the method proposed in this paper can mine the effective information in the knowledge graph more reasonably and effectively,thereby improving the accuracy.By mapping the vector to the unitary space,the relationship between nodes can be more accurately expressed in the recommendation application to improve the accuracy of the recommendation.Finally,this paper introduces the adaptive sampling method of reinforcement learning to fully excavate the global information in the graph,so that the knowledge graph information can be used better,and then the recommendation accuracy can be improved.