Graph Embedding Is Used To Classify Nodes In An Associated Network

In our real life,network or graph data is everywhere,and there are network data of different sizes in our social network and transportation network.In recent years,with the rapid development of the Internet industry,whether it is financial banking or electric commerce,the massive user data stored in it is enough to build a huge network system,and how to fully exploit the characteristics of nodes in the network system to help the users of the enterprise to carry out all-round Customer portraits and consumer behavior analysis are also hotspots and difficulties in this year's research.Based on the existing network feature learning algorithm and machine learning classification algorithm,this paper classifies and summarizes the traditional network feature learning algorithm and the emerging and more advanced network feature learning algorithms in recent years.Inspired by the literature,the overall network feature learning algorithm and machine learning algorithm are combined with the main classification learning task framework for multi-label nodes.The network feature learning algorithm in the framework includes matrix-based decomposition(Laplace feature map,LINE).Methods and methods based on random walk(deepwalk,node2vec);machine learning algorithms in the framework include logistic regression and autoencoder deep learning framework.This paper firstly proposes to embed the node feature vector learned by the network feature learning algorithm into the autoencoder framework for single task learning and multitask learning.The so-called single-task learning is supervised multi-label node classification,and multi-task learning is simultaneous unsupervised feature reconstruction and supervised multi-label node classification learning.In the improvement of the model,the node feature vector learned by the random walk algorithm is spliced into the adjacent vector of the corresponding node and then input into the autoencoder for multi-task learning.In order to verify the idea and effect of the combination framework of the proposed algorithm,three different types of datasets are selected,namely social network dataset(Blogcatalog and Youtube),biological network dataset(PPI),and word co-occurrence network dataset(Wikipedia).).Experiment to set different training sets and verification set ratios(7:3 and 8:2),calculate the scores of different network node feature learning algorithms on different machine learning classification algorithms(macro-F1 score),select appropriate hyperparameter combinations It is confirmed that the node2 vec algorithm combined with autoencoder and multi-task learning simultaneously has higher accuracy for the classification of the main task of the node than the other algorithms proposed in the paper.At the same time,in the improvement part of the model,it is also confirmed that the node node vector learned by node2 vec after the parameterization and the corresponding node are spliced as the input of autoencoder for multi-task learning,the node classification effect of the main task is compared with the mentioned in the text.Other frameworks are more accurate.This shows that the proposed algorithm framework has certain reference value in the classification of nodes based on actual services.