Mining Social Network Graphs Through Structural Clustering

Today's social networks are no longer the content of sociological research in a narrow sense.It has become a fiery research topic that integrates cutting-edge research value and huge commercial potential,attracting more and more researchers in various fields.With the development of the research,the data in the Internet has also increased rapidly with the speed of blowouts.The network in the era of big data has become extremely complicated.With the gradual in-depth study of the physical and mathematical characteristics of complex networks,researchers have discovered that in addition to the characteristics of small-world and scale-free nature,many real-world networks have a common feature,namely community structure.The community consists of a series of points and edges.It has the characteristics that the nodes within the community are closely connected,and the nodes between the communities are loosely connected.From the perspective of the community,it is better to mine the functions and values of the network,and it is easier to analyze the structure of the network and the characteristics of the network.Therefore,it is of great significance to mine the community structure in a complex network.Due to the lack of effective ways to convert social networks into data,the weighted and unweighted networks are treated as separate networks.Most algorithms for unweighted networks cannot be pushed to the weighted network.Based on this,this thesis mainly studies the way that social networks are transformed into data,which makes many clustering methods applied to data can be applied to social networks.This thesis first briefly describes the background of the dissertation,current research status,and the organizational structure of this dissertation.Secondly,this thesis expounds the meaning,related features,topology model and community structure of complex networks,and describes several typical community detection algorithms.Based on previous theoretical research work,This thesis proposes the concept of pseudo-adjacency matrix.Each row of the pseudo-adjacency matrix represents one node.On the diagonal elements of the pseudo adjacency matrix,we set the alpha parameter.This parameter ensures the accuracy of the social network into data,such that the data can better represent the structure of the social network.Based on the data,this thesis introduces the application of K-means,hierarchical clustering and FCM on social networks.In order to make the K-means algorithm better applied to social networks,this thesis proposes an initial value selection method based on the maximum node degree.This thesis uses the modularity,the standardized mutual information,and community in a strong sense and in a weak sense as indicators to verify the feasibility of the data conversion method.Through a large number of comparative experiments,we found that this transformation method has certain advantages.To a certain extent,the time efficiency and accuracy of community demarcation are improved.This method of transformation can be applied to weighted and unweighted networks,which improves the generalization ability of K-means and hierarchical clustering applied to social networks.