Study On Complex Network Community Discovery Based On DeepWalk Algorithm

Data,as one of the carriers of information,its importance is self-evident in this era of information explosion.Graph structure data is a kind of description method for complex networks.Its research value lies in the abstractness and generality about many phenomena in real world of complex networks.Improving the ability to analyze graph structure data can better explore various important characteristics of complex networks and analyze these complex networks existing in real worlds.Community structure is one of the important characteristics of complex networks,and it is also a characteristic of many complex networks in the real world.Its practical applications include,but are not limited to,black industry detection,precision advertising,and social network analysis.With the increasing data size and increasing data complexity,and in order to perform community discovery tasks on complex network data more effectively,this dissertation is based on the two modules of the Deep Walk algorithm in graph embedding,RandomWalk and SkipGram,making corresponding improvements for different network data,which can improve the effect of the node vector representation obtained by the Deep Walk algorithm in the community discovery algorithm K-means and effectively avoids the dimensional disaster.The main contents and contributions of this dissertation are as follows:1.In the case where only the adjacency matrix of complex network data is provided,combined with the conclusion of the PSO model,and based on the hyperbolic space coordinates of the nodes obtained by the Hyper Map algorithm,we define the weight calculation formula of the connecting edges between nodes that has evolved from the hyperbolic distance formula,and the hyperparameters are introduced to improve the generalization ability of the algorithm.By improving RandomWalk to Weighted Walk,we can extract the similarity between nodes and the popularity of nodes more effectively.Experiments show that among the five real network data selected,Weighted Walk can indeed improve the community discovery effect compared to RandomWalk in the native Deep Walk algorithm.2.On the basis of providing topological information of complex network data in real network research,it also provides the node's own attribute information.Based on these node's own attribute information,the corresponding function is set to obtain the node's similarity matrix.By using the similarity matrix,RandomWalk can be improved to Weighted Walk,which can help improve the performance of Deep Walk algorithm.3.Based on the above-mentioned node similarity matrix,the objective function of the SkipGram model is studied and improved,the embedded vector update formula of the nodes is improved during the training process,and the similarity information is incorporated into the iterative process.Research shows that in real network data,the improved SkipGram model has a significant improvement over the native SkipGram model,and the effectiveness of the algorithm is verified through simulations.