Research On Key Technologies Of Community Detection In Online Social Networks

In recent years,the rapid development of communication network technology has brought great convenience to people's daily life and work.Especially,the rising popularity of online social network makes millions of internet users more easily to be working,studying,socializing and entertainment online.Through these "online" interactions,people can not only form variety social relations in real life,but also build a rich variety of new social relations in the online networks.The emergence of online social network has caused the wide attention of the researchers.Community is the basic building blocks of network social relations,detecting communities are of great importance in computer science biology and sociology networks.Community detection in the online social networks,will help us to reveal the structure and features of online networks,and understand the operation mode and complex behavior of networks,which has very great research significance.Online social networks usually have properties of diversification,dynamic changes and large scale,which have presented new problems and challenges to community detection.Traditional community detection methods of elemental,static and global are difficult to effectively deal with these processing requirements.New community detection models and methods need to be developed.This topic belongs to the online data mining scope,which relies on the national science and technology support project " public opinion monitoring over integrated network " and the national 863 plan project " users relationship and behavior analysis technique in the public communication network ",aiming at breaking through the key technology of online community detection in social networks and providing technical support and services for the online social network information processing and data management.In summary,the main work and achievements of this dissertation can be summarized as follows:? The application background of research topic in this desertion is introduced,and the classical models and methods of community detection research are summarized and teased out.Then,an intensive analysis of online social networks' new characteristics is present,and pointed out the key problems of our subject faced.Lastly,the research scope and the technical route of this article are specified,and the new models and solutions of community detection are explored.? When detecting communities in social networks,there are two possible sources of information: the networks' structure attribute and the nodes' content attribute.But the effective fusion methods are lacked.Based on this,this dissertation proposes a community detection method based on Joint matrix decomposition(CDJMF).Our method based on the assumption that from the two different information sources of linkage and node attributes can get an identical nodes' affiliation matrix.This method firstly calculates the node similarity to construct similarity matrix according to the content of property.Then,joint matrix decomposition model is constructed by combining with the feature of topology structure,and these two sources of information are present by collaborative learning of community detection.At last,through the iterative decomposition of joint property space matrix,the community detection results are given.This method can effectively combine different two information sources,and improve the effectiveness and robustness of the community detection in multi-attribute networks.Experimental results show that our method can effectively detect network community through variety of attribute information,and has a higher community division quality.? How to effectively combine the network structures on different time points is the key and difficulty to affect the performance of community detection method in dynamic networks.Based on this,a semi-supervised dynamic community algorithm based on non-negative matrix factorization(SDCD)is proposed.Firstly,our method extracts the effective stable structural unit contained in the historic moment network.Then,the stable structural unit is used as a regularization supervision item to guide the community detection in current moment of network society.Our method can effectively use the information of stability structure contained in the historical moment,and analysis integration information on both historical and current moment in the same architecture,which provides a new research framework for dynamic network community detection.Experiments on the real network data sets show that the proposed method can effectively use historical information to guide the community detection in current moment,and can more accurately reflect the variation of community structure during the network evolution.? Aiming at the dynamic real-time processing requirements of the online social network,a new method for the community detection in flow graphs based on online non negative matrix factorization(ONMF)is proposed.Firstly,our method puts graph data into the cache as continuous streams to deal with.Then,our method iterative updates the existing community belonging matrix real-time using online nonnegative matrix decomposition architecture and by means of the projected gradient descent theory.Lastly,through effective learning rate and cache strategy setting,our method ensures the convergence and rationality of graph stream processing.Experiments on real network data sets show that methods based on ONM has a higher community detection quality compared with existing methods,which can effectively deal with steam data in the real time.? Local community detection in large-scale network have a number of limits,such as the detection results are sensitive to the position of source node,the topology information is difficult to be used effectively.In this paper,a local community detection method based on local influential nodes set(IN-LCD)is proposed.Firstly,the local influence index of the node is defined,and the influential nodes set near the source node is calculated and constructed.Then,from the influential node set,the continuous expansion of the community is realized.Finally,through the calculation of the similarity index between nodes and community,the whole local community is constructed.The method can effectively overcome the problem of local community detection in the initial node position sensitive problem by using the most influential node set to expand the community.Experiments on both real and artificial network data sets show that the recognition performance of IN-LCD is better than the existing local community detection method,which is more effectively in utilizing the local information in large scale networks.