Research On Link Prediction Based On Network High-Order Subgraph Structure

Complex networks as an important tool for studying complex real systems have become a hotspot in scientific research.Link prediction,as one of the important research directions in the field of complex networks,has attracted the attention of more and more researchers.Link prediction algorithm based on the similarity between nodes is one of the more popular research directions in the field of link prediction.The main idea of this algorithm is that the higher the similarity between two nodes,the higher the probability that the two nodes have links..As one of its research directions,link prediction algorithms based on network topology information have gradually become a research hotspot due to the availability of network topology.In addition,the method based on network node representation can also be used in link prediction.Here,the node representation is to use the existing information in the complex network to build a node representation model to automatically learn the low-dimensional vector representation of the node.However,most of the existing work in traditional link prediction algorithms based on network topology information is based on predicting the common neighbors of the two target nodes,instead of treating the local topology information of the predicted node pairs as a whole,thus ignoring a lot High-order topology information that is valuable for improving link prediction.Based on the analysis of existing algorithms,this paper proposes a link prediction method based on weighted surrounding subgraphs.This method mainly learns the subgraph pattern that promotes links based on the local surrounding subgraph structure of target nodes.Compared with the traditional network topology-based link prediction algorithm,the weighted surrounding subgraph-based link prediction takes the target node to the surrounding subgraph as a whole to build a prediction model,so that the constructed link prediction method contains more valuable topological information.In the link prediction method based on weighted surrounding subgraphs,how to use the leaning machine to learn the surrounding subgraph structure of the target node pair is the key to obtaining subgraph pattern promoting link.This paper proposes a link prediction method based on the LSTM learning subgraph structure.The surrounding subgraph of the target nodes is taken as a whole,and then the obtained surrounding subgraph is mapped to a meaningful node that the LSTM can learn using the node encoding algorithm.And the node embedding is used to as the input of LSTM.Experiments have showed that the proposed link prediction method in this paper improves the effect of link prediction.