Overlapping Community Detection Based On Deepwalk And Graph Neural Network

With the rapid development of Internet technology,many systems in the nature can be abstracted as complex networks.How to accurately and effectively detect overlapping communities and quickly realize functional division is a problem in the field of complex networks.At present,complex networks can be divided into no attributes complex networks and complex networks with attributes.Overlapping community detection algorithms for complex networks without attributes are mostly based on structure division,but the accuracy and stability of these algorithms need to be improved.Some of the attributes complex networks overlapping community detection algorithms ignore attributes information and have a large information loss.Although some algorithms make full use of structures and attributes information,they have relatively large time and spatial overhead.To solve the above limitations,this paper studies the overlapping community detection algorithms based on the Deep Walk model and graph neural network.Firstly,in order to solve the problems of poor stability and low accuracy of complex networks without attributes,this paper proposes Deep Walk-based overlapping community detection algorithm.This paper uses the Deep Walk model to learn the network's topology to obtain low-dimensional vector that reflect the spatial location of nodes and constructs the weight matrix through vector dot product operation.Through the label propagation algorithm with a preference selection strategy,the neighbor node with the highest preference probability is updated on the basis of preserving the node's own label.When the termination condition of iteration is reached,labels are uniformly divided into the same community to obtain stable overlapping communities.Secondly,it discusses the large information loss,high time complexity and spatial complexity of the overlapping community detection algorithms of attributes complex networks.This paper proposes an attributes complex network overlapping community detection algorithm based on graph neural network.The algorithm builds a graph neural network model with an adaptive attention mechanism,combines the structure of the complex network with attribute information,and uses the adaptive attention mechanism to aggregate neighbor nodes information according to different influences to obtain the nodecommunity membership strength matrix.By minimizing the negative log-likelihood function of Bernoulli–Poisson model to train graph neural network model parameters.when the iteration is stopped,the model parameters and the node-community membership strength matrix are optimized,and the optimal matrix is mapped to obtain a stable overlapping communities.Finally,the proposed two algorithms are applied to real-world datasets and Synthetic datasets respectively.The effectiveness of the proposed algorithm is verified by combining the evaluation indicators analysis with the baseline algorithms.