Research On Nodes Influence In Social Networks

The rapid development of Internet technology, the humanization andsocialization of network applications promote the exchange and communicationbetween people, speed up the social network’s prosperity and progress. Social networkcontains rich information of user and link relation. From the perspective of datamining, social network is represented by graph with multiple relationship of data sets,among which a node in the graph represents an object, an edge represents relationshipbetween objects.In the past few decades, social network is paid more and more attention. Socialnetwork analysis and research, in-depth excavation of the network structuralcharacteristics, group behavior and the transmission mechanism have practicalsignificance to know more about network system. At present, the questions about howto find community structure of the network and how to evaluate the importance of thenetwork users effectively are the focuses in social networks research.Community is one of the most important properties in social networks. It is anobject’s group with common nature. The objects in the same community connecttightly but relatively sparsely between communities. Community detection is toidentify the closely related network node sets, which can be understood as a subgraphrecognition. It is an important task to mine the potential structure in social network,which has attracted more and more researchers’ attention in the field of data mining.In addition, mining important nodes in the network is another important direction insocial network analysis. It has a widespread application and meaning to assess theinfluence of nodes reasonably for the social management, business, marketing and soon.On the basis of research and analysis about the traditional and moderncommunity detection algorithms, a new community detection algorithm based onrepresentative node is proposed-the RCD algorithm. The RCD algorithm adopts theidea of CURE clustering algorithm, proposes using multiple representative node torepresent a community, using eccentricity as a basis for the selection of center node,then selecting representative nodes by the center node according to the similarity of nodes, then according to the similarity of the communities, merging the communitieswith the bigger similarity in turn until reaching the desired number of community,finally we get the community structure of tight connection. When calculating thesimilarity of nodes, we correct the limitations of Jaccard similarity to make thesimilarity calculation more reasonable. Algorithm experiments on karate data set,American College football data set and co-authorship data set in ACM SIGMODfrom DBLP, and compares with several classic algorithms. It verifies the validity andflexibility of RCD algorithm.In addition, this paper proposed an influence assessment method based on node’sconnection mode on the basis of community detection. The method is different fromthe traditional node influence assessment methods. The latter assesses mostly from thetopology of the network as a whole, but the former divides the node’s connectionmode in the community into two categories from the aspect of local communityinformation network. It is more important for the nodes that are all connected withdifferent communities. Because it is equivalent to the role of a bridge between thecommunities and play a key role for the communication between communities.Algorithm validates on the three data sets, and conduct Spearman consistency contrastanalysis with degree centrality and betweenness centrality. It verifies the reliabilityand rationality of the evaluation method based on the influence of the node’sconnection mode.In conclusion, this paper can not only improve the community detection qualityof social network, but also makes the evaluation of the network nodes more effective.The main focus on the research in the future is to improve the efficiency of thealgorithm and ensure the accuracy as well to adapt to the large-scale network. Inaddition, taking more information about the nodes into consideration to make theassessment method more reasonable.