A Study Of Feature Analysis And Community Mining In Social Networks

The relations among the members of society are usually described by socialnetworks. With the rapid development of computer network technology, the emergingsocial networks based on Internet such as E-mail networks, peer-to-peer networks,social networks and blog networks develope greatly, and give a profound influence onthe behavior patterns of the human society. Therefore, research on the social networks ismeaningful both in theory and in practice. The social networks are often large-scale andcomplex in structure, therefore complex networks theories and models are adopted tostudy the social networks.Real social networks usually have some structural features of the complexnetworks, such as community structure, the scale-free degree distribution, clustering,"small world" network, dynamic evolution and so on. The community structure is animportant structural feature of social networks. Community detecting of the complexsocial networks is the focus of the social networks study in recent years. Modularity hasplayed a very important role in community detecting. It has been used to evaluate thequality of community structure detecting. In addition, the social networks consume alarge portion of traffic on the Internet; therefore the social networks measurement is alsovery important. It can provide an insight into the topology characteristics of the socialnetworks, monitor network traffic, and ensure network security. The main research workof the thesis is mining the community structure of the social networks based on complexnetwork theories, analyzing the structural characteristics, and monitoring the socialnetworks actively. Through those works, the evolution of the real social networks isrevealed. The innovative achievements of this thesis are as follows:(1) A community detecting method of the static social networks based on thedirected and weighted modularity is proposed.Based on the unity of the undirected/directed networks and theunweighted/weighted networks, a new community detecting method SNCD of the staticsocial networks is proposed using the new modularity optimization method. Byintroducing the edge direction and weight, the directed and weighted modularity is used to detect the community structure. The accuracy and effectiveness of communitydetecting dramatically improved with much more available information utilized. Themethod uses the heap structure and multi-tasking modular architecture, which increasesthe computational efficiency greatly. The method can worked efficiently on large scalenetworks which have millions of nodes. At the same time, the dividing accuracy of themethod improves largely comparing to the typical methods widely used currently.(2) A community detecting model of the dynamic social networks based on timeseries is designed.Time series is used to describe the dynamic and the static characters of the socialnetworks. That is, the social networks are cross-sectionally static but evolutedynamically on whole time series. By introducing a new modularity definition based onstructural similarity, a new community detecting model of the dynamic social networksbased on time series is designed. The model first proposes a static community detectingmethod LMA to detect the static community set for every moment of the time series,and the same time provides the intermediate community set for next DNCD method.Then by extending the LMA method, DNCD method is proposed to detect dynamiccommunity of the dynamic social networks based on time series. After that, the DNCDmethod combines the intermediate community set with the community time-sequencechain, and obtains the evolution trajectory of community structure and the final stablenetwork structure.(3) A social network active measurement strategy is proposed.To better understanding the structural features of the social networks andmonitoring the real-time status of the networks, a multi-protocol and modular activemeasurement strategy for the social networks is proposed. Using the BitTorrent protocolof peer-to-peer network as the measurement subject, the active measurement strategy isadopted, and the measurement indexes are designed. The real-time measurement of thereal social networks is realized, and the inherent structural characteristics and behavioralpatterns of the real social networks are revealed. Taking the BitTorrent protocol ofpeer-to-peer network as measuring subject, and by designing corresponding measuringindexes and adopting active measuring strategies, we realize real-time measuring of thereal social network and reveal the inherent structure and behavior patterns of the realsocial network.