Analysis And Prediction Of Community Evolution Based On Network Representation Learning

In recent years,the evolution analysis and prediction of communities has become the focus of research on dynamic complex networks.Tracking and predicting the path of the evolution of the community is of great value in grasping the network situation and predicting the trend of the spread of diseases.In the current research on community evolution analysis,there is a lack of consideration of network topology,which makes it difficult to ensure the accuracy of community evolution event identification results.However,in the community evolution prediction research,there are problems that the extracted feature set is not comprehensive enough and lacks the characteristics that can describe the internal topological structure of the community.Aiming at the limitations of the above community evolution research,combined with the idea of network representation learning,researches are carried out in two aspects: community evolution event recognition and evolution prediction.The main work is as follows:Firstly,a community evolution research method based on node influence and multi-attribute fusion is proposed.In this method,the idea of network representation learning is integrated into the process of node influence calculation,and a node influence calculation method based on network topological structure is proposed.Evaluate community differences through node influence and integrate community stationarity and difference to measure the similarity between communities.At the same time,based on the community similarity measurement method,a community evolution event detection model EMNI based on node influence is proposed.The model redefines the discriminative methods of seven kinds of community evolution events: forming,dissolving,growing,shrinking,splitting,merging and continuing,and to identify community evolution events that are more in line with the real network situation.Secondly,a research method for community evolution prediction based on multivariate feature sets and latent structural features MF-PSF is proposed.The method constructs a multivariate feature set for community evolution prediction from four aspects of community core node features,community structure features,community time series features,community behavior features,and combine the idea of network representation learning to extract the potential structural feature of the community.The potential structural features of the community obtained through Deep Walk and the idea of spectral propagation fully consider the overall structural features of the community and the distribution of community vertices,combine the constructed multivariate feature set as a feature of community evolution to improve the accuracy of community evolution prediction.Finally,experiments are carried out on two real data sets,and compared with other models,it is verified that the EMNI model proposed in this paper can identify more community evolution events,and the distribution of various events is more balanced,which is more in line with the evolution of real networks condition.At the same time,the proposed method MF-PSF for community evolution prediction based on multivariate feature sets and latent structural features is verified on different datasets,and the importance of each feature in the evolution prediction process is analyzed.The experimental results show that the potential structural feature of the community extracted in this paper is effective,and combine with the constructed multivariate feature set can effectively improve the accuracy of evolution prediction.