The Research On Link Prediction In Knowledge Graphs With Network Structural Features Integration

Link prediction aims to study the mechanisms for forming connections between nodes that are not yet linked in complex networks and knowledge graphs.With the deepening of network science research,studying link prediction problems from the perspective of network structure has received increasing attention.In light of this,this paper proposes a method for testing link prediction based on network structure in the context of complex networks to verify the effectiveness of network structure in link prediction problems,and then applies network structure to link prediction problems in knowledge graphs to address graph completion problems.The main contributions of this paper are as follows:(1)A resource broadcast-based complex network link prediction method is proposed.This method integrates two key factors in resource transmission,namely transmission paths and resource loss,and introduces the concept of resource broadcasting.Resource broadcasting considers bidirectional transmission of resources and four typical transmission paths,and evaluates possible resource loss during transmission.Node similarity is characterized by the amount of resources transmitted between nodes,and link prediction is performed based on this similarity.Experimental analysis conducted on eight real-world networks compared to baseline methods shows that this method has better prediction accuracy.(2)A knowledge graph link prediction method that integrates network structure is proposed.This method considers the graph structure formed by the factual triplets on the basis of the transfer model.First,neighboring entities are mapped to the structural space based on the central entity’s adjacency relationships,forming the central entity’s structural vector.Then,the structural vector and the semantic space entity vector are fused to model the entity’s distributed representation.In link prediction tasks,this model shows improved knowledge representation capabilities compared to baseline models,revealing potential facts in the knowledge graph and improving the accuracy of graph completion.