Research On Community Detection And Prediction Algorithms On Dynamic Networks

With the continuous development of the Internet,the connection between people is getting closer,resulting in the weakening of individual consciousness and the enhancement of team capabilities,and gradually forming a huge social network based on communities with different attributes.Social networks often show that the internal relations of the community are very close,and the relationship between the community is very sparse.Community detection is the process of reversing the formation of communities,and identifying closely connected community structures from complex social networks.However,the structure of social networks in real society is not fixed,it is constantly changing with time.Community detection based on dynamic networks and prediction of community evolution are of great significance in studying the structure of social networks.Most of the traditional community detection methods involve static snapshots of the network,which do not reflect the dynamic changes of the community in multiple snapshots.The main content of this study is to incorporate the dynamic changes in the community into the analysis to predict the trend of network evolution,combine the topology of the nodes in the dynamic network,and the text information,improve the effectiveness of community detection based on representation learning,and increase the time analysis enhancement The effect of community evolution prediction.First,we study the problem of community detection in dynamic networks,and propose a new multi-representation learning algorithm DGAE(Dynamic graph Auto-encoder)based on graph convolutional networks.In a continuous time-series,embed user relationships and text content under multiple timestamps in a dynamic network into a Temporal Shared feature matrix,and learn users associated with text information by using graph convolutional networks Potential characterization.Combine the user's various information to improve the accuracy of community detection,and finally obtain the community network structure of the dynamic network with the required timestamp through clustering.Secondly,we study community prediction in dynamic networks,and propose a kind of deep self-coding network TDAE(Temple Deep Auto-encoder)with time analysis.The Shared matrix under multiple timestamps in the dynamic network was re-embedded into the Temporal Shared feature matrix.Through refactoring depth since the encoding learning,thus obtain contain community evolution potential representation of information,obtain the potential characterization in the study of the features of the network and the continuous changes in network time evolution characteristics,the decoding of predict the evolution of network can be used as a community,the community structure in the future is acquired by clustering the final prediction results.In this paper,two Synthetic dataset and one ar Xiv dataset were used as data sets to verify the algorithm effect,and the DGAE and TDAE algorithms were evaluated through the two indexes of Jaccard(similarity coefficient)and Purity(purity coefficient).Experiments show that,compared with the community detection benchmark algorithm,the DGAE algorithm proposed in this paper has higher accuracy and less time consumption.The TDAE algorithm proposed in this paper is compared with the community detection benchmark algorithm.The TDAE algorithm has an effective community evolution prediction.Compared with the community prediction benchmark algorithm,TDAE algorithm has better accuracy.