Identification Of Opinion Leaders In Social Networks Based On Overlapping Community Structure

With the development of Internet and modern communication technology,social network has built a huge platform.With the help of social network,people can share and exchange information with others in anytime and anywhere.The research on social network has two classical issues,which has a very important practical significance for understanding the structure and discovering the hidden rules of social network,and provides guidance for the use and improvement of the network,that is community detection of nodes and identification of opinion leaders.So far,many scholars have proposed a series of algorithms for community detection.Although most of the community discovery algorithms can effectively recognize the community structure of the network,there are still some shortcomings such as high parameter dependency and unreasonable selection of seed nodes.There are also many algorithms to identify opinion leaders,such as algorithms based on local characteristics of nodes,algorithms based on topology of the whole network,and so on.However,these methods usually only rely on the individual attributes of the nodes to measure the influence of nodes,ignoring some information of node which can not accurately reflect the importance of node in social network.Based on the above analysis,this thesis mainly studies and improves overlapping community discovery algorithm and opinion leaders detection algorithm in social network.The main reaearches are as follows:(1)An overlapping community discovery algorithm is proposed based on three-stage strategy(Three-stage strategy,TSS).In order to make the seed nodes distributed throughout the network,the thesis firstly selects the nodes with large influence as the seed nodes according to the similarity values between the nodes in the network.Secondly,the initial communities are constructed by selecting seed nodes,and the initial communities are expanded by adding or deleting nodes selectively according to the threshold and membership relationship.And then,the nodes are selectively added to the initial communities to form a larger initial community by comparing the membership and threshold between the initial neighbor nodes and the initial communities.Later,by synthetically evaluating the score of the nodes,the overlapping communities are optimized according to the local gravitation function to get the final community structure.Finally,the algorithm is verified on the discrete dataset and text dataset that the algorithm can efectively identify overlapping communities in the network and the quality of the partition is good.(2)An opinion leaders detection algorithm is proposed based on overlapping community structure,which called Community-based Mediator method(CMB).In view of the problem that most opinion leaders recognition algorithms only consider issues such as a single network structure or a single attribute of a node,the thesis comprehensively takes into account the internal and external densities of nodes in the network.The internal and external densities of nodes include the strength of internal and external edge links,the strength of internal and external location links,and the strength of internal and external distance links.Meanwhile,multiple attributes of the internal and external density of the nodes are comprehensively evaluated and the influence weight of multiple attributes is calculated respectively.Later,the influence of nodes is evaluated according to defined influenced function and the users are ranked according to the value of influence function.And then,high influenced users are selected as opinion leaders.Finally,the algorithm is verified on the discrete dataset and text dataset that the algorithm is effectiveness and the opinion leaders are accurately identified.