Application Research On Essential Nodes Search And Link Prediction In Complex Networks

With the continuous development of network science and the growing size of information data,the scales of the networks are increasing tremendously and the studies of large-scale networks have been an upsurge of research.In view of the advantages of representation learning algorithms for researching large-scale network,the traditional researches based on network knowledge,like essential nodes classification and link prediction,have begun to investigate with representation learning algorithms,which have achieved significant results.Combined the network science knowledge with the representation learning algorithms,we propose the algorithms for essential proteins classification and link prediction.Firstly,we propose a method for the essential proteins classification that utilizes knowledge of biological information.In the research of searching essential nodes in networks,many experiments have proved that the methods of combining multi-sources information are more effective than the methods of considering only a single kind of knowledge.However,the existing search methods do not fully consider the knowledge contained in the network itself,and many key information are lost.The essential proteins classification method proposed in the article is integrating the biological information of the protein nodes in the PPI network embodied in the STRING database.At the same time,the representation learning algorithms are used to extract the topological structural features and biological information features of the protein nodes in the network to implement the classification of essential protein nodes.By analyzing the experimental results,the accuracy,the recall and F1-value of the essential proteins classification algorithms are higher than that of the comparison algorithms,thus demonstrating the effectiveness and advancement of the algorithm.Secondly,we propose a link prediction algorithm based on the Probase database.Link prediction is to explore similar nodes in the network by analyzing the network structure and the attributes of the nodes,and further predict the nodes that are potentially connected with the known nodes.The link prediction method proposed in this study mainly utilizes the representation learning algorithm to vectorize the network and determine the degree of similarity between nodes based on the similarity of node vectors.By computing the hit ratio of top-k results and the similarity between the predicted node and the given node,the stability of the representation learning algorithm on the link prediction task has been proved.In summary,the employment of multi-sources information combined with representation learning algorithm can effectively improve the accuracy of classification of essential protein nodes in the network.At the same time,relying on the representation learning algorithm to vectorize the network,link prediction can be conducted by calculating the similarity of nodes,which can improve the prediction hit rate and ensure the stability of the prediction.