Research On Dynamic Network Community Discovery Algorithm Based On Network Representation Learning

All kinds of relationships that exist in the real world can be abstracted into a network,and all of them have a common characteristic,namely,community structure.It truly reflects the various characteristics behind the network structure,such as potential interest groups in social networks,disciplinary relevance in citation networks,and potential functional modules in protein networks.Such networks are changing all the time,and the study of dynamic networks helps to better analyze and predict the behavior of individuals in the network,realize accurate group promotion,and significantly reduce marketing costs;discover and explore potential correlations,realize efficient target search,and improve process efficiency;understand and dig unknown functional modules,reveal the essential laws,and promote long-term development in multiple fields.With the increasing scale of dynamic networks,the network representation and community discovery of dynamic networks face two major problems that need to be solved urgently.On the one hand,the low quality of obtaining various data features of graph structure leads to the bias of dynamic network representation,which makes the community discovery algorithm on this basis not ideal;on the other hand,facing the large-scale dynamic changing network,there is the problem of unstable community discovery results brought by the random selection of node update order and label update strategy.In response to the above problems,In response to the above problems,a dynamic network-oriented representation learning model(Dyn NE)is proposed,which extracts the topological structure features of the dynamic network through graph convolutional neural networks,uses the set aggregation function to aggregate the neighbor information of network nodes,and adopts an improved GRU model combining timerelated features on the network,enriching various feature information of network nodes,forming a network representation learning model that truly reflects the network structure and its dynamic changes.In addition,a dynamic network community discovery algorithm(TSLPA)based on label propagation is proposed.Through detailed analysis of network node capabilities,SRank algorithm is used to mine potential core nodes in the network,and the initial community is built based on this.The two-stage community discovery algorithm uses incremental community optimization update strategies to effectively reduce the accumulation of errors and the formation of black hole communities.The experimental analysis results on three different data sets show that the Dyn NE network representation learning model has significantly improved the F1 value and average accuracy of node classification and link prediction tasks,respectively;the TSLPA community found that the algorithm is relatively high compared to the benchmark algorithm.In other words,whether it is in terms of modularity,ARI and NMI,etc.indicators have been effectively improved.