Research On Knowledge Graph Link Prediction Based On Subgraph Reasoning

As a structured semantic representation,knowledge graph can models entities,concepts,attributes,and their relationships in the real world.Knowledge graph reasoning aims to deduce new relationships between entities from existing data through knowledge reasoning.Knowledge graph reasoning plays a critical role in many downstream tasks,such as question answering systems,recommendation systems,search engines.In view of the broad application prospects of knowledge graphs,the study of knowledge reasoning on large-scale knowledge graphs has become a research hotspot in the field of natural language processing in recent years.At present,many researches solve the link prediction problem between entities by mapping entities and relations into a vector space or searching the paths between entities.Although these methods can acquire relational features in the knowledge graph,they only consider the impact of a single path or first-order adjacent information on relational reasoning.Not only do they ignore the more complex and high-order interrelationships that exist widely between entities,but they also fail to synthesize the rich information of multiple paths,and the capture of the characteristics of relationships between entities is not comprehensive.Therefore,this paper fully considers the rich information contained in multiple paths,proposes a novel subgraph-based knowledge graph relational reasoning method Sub GLP.By constructing multiple paths into a subgraph,and then using subgraph reasoning to predict the relationship between entities,it can not only improve the reasoning ability,but also alleviate the error accumulation problem of path reasoning and improve the interpretability of relational reasoning.So as to complete stable and efficient relational reasoning.Sub GLP first extend the paths between entities to subgraphs,then combine the graph embedding representation with the graph neural network to calculate the subgraph features.Finally,it calculate the neighborhood structure information of entity pairs from the subgraph structure to conduct the link prediction between entities.Second,aiming at the problem of insufficient correlation between subgraphs of node subgraphs and subgraphs in Sub GLP,the paper change node subgraphs and edge subgraphs from parallel to serial.On the basis of Sub GLP,propose link prediction method based on hierarchical subgraph reasoning in knowledge graph.In addition,add the attention layer to assist in entity-to-relationship prediction.Extensive experiments are conducted on two commonly used knowledge graph inference benchmark datasets.The experimental results show that the Sub GLP model proposed in this paper outperforms the existing benchmark methods,and the improved HSub GLP has further improved in the accuracy of relational reasoning.