Research On Semi-supervised Community Detection Methods Integrating With Attribute Information

Many complex systems in the real world can be abstracted into complex networks,and community structure is an important property of complex networks.Studying the community structure is of great significance for understanding the topology and functional characteristics of complex networks.The real network structure is usually sparse,and the community structure of some networks is not clear enough.Most of the existing community detection algorithms only consider the network topology,and have limitations when dealing with sparse networks and fuzzy communities.Node attributes and semi-supervised information that are underutilized but widely existed in the network can effectively compensate for the lack of structural information.Therefore,this paper combines node content and semi-supervised information to study the community detection algorithm.Main tasks are as follows:Firstly,a semi-supervised community detection algorithm integrating with node attributes(SCDAN)is proposed.Based on the non-negative matrix factorization model,the network topology,node attributes and semi-supervised information are effectively integrated into a unified objective function.So the structure and content information complement each other,prior and attribute information complement each other as well.Secondly,the update rule corresponding to the objective function is derived,and the KKT condition is used to optimize the objective function,and then the more accurate community detection results are obtained.Experiments have been carried out to verify that the network structure,node attributes and semi-supervised information all have improved the detection results of the community.Finally,the model is validated on the real data sets and compared with various comparison algorithms to prove the superiority of SCDAN algorithm in the accuracy and stability of the community detection.Moreover,the framework can obtain the specific attributes of the community and explain the true semantics of the community.In addition,the influence of the proportion of individual labels on the detection results of the community was studied,and the sensitivity analysis of the experimental parameters was carried out.Finally,the Aminer dataset is used for empirical analysis,we extract the semantic information of the community and explore the research focus and research interests in several sub-areas of biology,which provides reference for researchers to carry out research activities.In summary,based on the non-negative matrix factorization model,this paper proposes a semi-supervised community detection algorithm integrating attribute information,which has high accuracy and stability,and can semantically interpret the community.