Research On Label Propagation Algorithm And Its Application In X Enterprise Security Risk Control

In recent years,with the gradual improvement of network communication infrastructure construction,the application of various new technologies and the improvement of computer computing power,a vast amount of data has been accumulated in various scenarios of human life,work,social interaction and shopping.At the same time,a large number of data analysis and data mining needs have emerged in various fields,and machine learning is gradually playing an important role in these fields.Machine learning is divided into supervised learning,unsupervised learning and semi-supervised learning according to the degree of dependence on label samples.Semi-supervised learning mainly USES a small number of labeled data and a large number of unlabeled data for joint training.Through the classification and clustering of data through the training model,it effectively solves the heavy statistical analysis of data and saves the cost of labelling data by a large number of people.Semi-supervised learning is a hot topic in machine learning.This paper mainly studies semi-supervised learning based on label propagation algorithm.In the traditional tag propagation algorithm,the data on the classification boundary is easily divided into other classes under the influence of edge weights when clustering samples.In this paper,a graph construction method based on sample clustering analysis is proposed on the basis of the traditional label propagation algorithm theory,and an optimization scheme of edge weight influence coefficient is added in the process of constructing the graph,which can effectively avoid the trap that samples at the classification boundary are divided into other classes by mistake and improve the accuracy of label propagation clustering.At the same time,combining with the X enterprise current situation,the optimized label propagation algorithm is applied to the X enterprise security risk prevention and control link,through the effective data clustering fast label goods together into clusters with similar risk,to the same reviewers to group quickly audit,improve the efficiency of personnel audit,reduce the human cost of the enterprise.Through the practical application in X enterprise,according to the sampling analysis of the actual results,it shows that the optimized algorithm design based on multi-factor edge weight is superior to the traditional classical tag propagation algorithm in most cases.