Evolution Analysis And Prediction Of Communities In Dynamic Social Networks

With the popularity of social networks,social media is generating massive amounts of data every day.Most social networks are real-time which change a lot every moment,so they are also known as dynamic networks.In order to dig the huge value hidden in the data,dynamic network analysis has become a research hotspot.Thereinto,community evolution analysis and community evolution prediction in dynamic networks are the most significant research orientations,which are of great value in application of influence analysis,information communication and network marketing.Based on the existing researches on community evolution evaluation criteria,evolution classification model and evolution prediction model in dynamic networks,the community evolution classification model and the feature extraction in prediction model are focused on research in this thesis.The main work includes:Firstly,an index of community evolution evaluation based on the node leadership is proposed.Above all,a method of assessing the leadership of nodes is proposed based on the characteristics of dynamic network,and it is introduced into the index of community evolution evaluation.It is used to evaluate whether the two communities are similar.This index emphasizes the influence of node weight on community evolution,and is more suitable for real dynamic networks.Secondly,a community evolution classification model based on node leadership is proposed.The criteria of community evolution evaluation based on the node leadership above is used as the basis of community evolution classification model.According to the evolution of the real networks,community evolution is divided into seven categories.The experimental results show that the model proposed can detect multiple types of evolution events better than the others,especially on Splitting and Merging evolution events.Furthermore,in order to gain a more intuitive understanding of community evolution,a novel Evolution-Tree model is proposed that clearly shows the entire life cycle of a community from its generation to disappearance.Thirdly,the future trends of communities are predicted based on the characteristics of community evolution,especially their historical evolution information.And three important laws are found out by focusing on the relationship between features and various evolution types.The laws are as follows: 1.Small communities are more likely to grow than big communities.2.A community which grew once or more times in its evolution history is likely to merge with other communities in the next time window.3.Large communities are more prone to dissolve than small communities.In the experimental results,both the Growing event and the Merging event show higher prediction accuracy,while for the Continuing and Shrinking events,the results are less favorable due to the less training sets.