Research On Mining Structural Hole Spanners In Weighted Networks

More and more researchers focus on how positions in social networks benefit those people who occupy them. The so-called intermediary or bridge is which exists between individuals or between communities, they tend to have access to the richer information and have more control on the relationships in the networks. The point of view is on the basis of the structural hole theory. The structural hole theory is an important conclusion to study the key position and key role in groups of sociology and social network, which interests many researchers of sociology, psychology and economics and the structural hole theory is widely used in different fields. As an important concept of network structure analysis, structural hole spanner plays a key role in terms of access to effective information of network.Few works investigate the closeness between nodes of weighted networks and mine the structural hole spanners. However, in real networks, the weight of edge is the key role that influences the network performance. As community is an important structure in networks, what’s more, the effective information generally diffuse from one community to another community. The users who diffuse the information between communities are same with the structural hole spanner. The dissertation focuses on how to mining the structural hole spanners in weighted networks and a framework based on community was proposed to mining structural hole spanners in weighted networks. Firstly, a community detection algorithm is used to detect the communities in weighted networks. Then the importance of each node in communities is initialized, and a couple of functions are updated to mining the structural hole spanners finally. For initializing, two methods are given to initialize the importance of the nodes in each community. For the first method, based on the structural hole theory, a user who connected to opinion leaders between different communities is more likely to be a structural hole spanner. For this intuition a method called W_HIS is proposed based on weighted PageRank for weighted networks. As for the second method, the Constraint Index is an important evaluation index of structural hole theory and the smaller Constraint Index can be regarded as structural hole spanners, thus a method referred as W_CIHIS which initializing nodes importance with Constraint Index is proposed to mining structural hole spanners in weighted networks. Experimental results on real datasets and public datasets show the efficiency of the proposed methodsThe main work of this dissertation is as follows:Firstly, based on the structural hole theory proposed by Ronald Burt, we describe the implication of structural hole, the research background and significance of structural hole in weighted networks. We briefly review the development and research status of structural hole theory. Then, we introduce and analysis the methods to measure the structural holes of researchers. Finally, the author proposes two methods to mining structural hole spanners for weighted networks.1. The first method is based on the intuition of structural hole spanners usually connected with opinion leaders:W_HIS. We first detect the communities of the network, and then improve the classical PageRank algorithm on weight. Initialize the node importance in each communities with the weighted PageRank algorithm, and a couple of functions are updated to mining the structural hole spanners finally.2. The second method is based on the weighted Constraint Index:W_CIHIS. The author studies Burt’s Constraint Index and improves it with weight. After detecting communities of network, we initialize the node importance in each communities with the weighted Constraint Index, and a couple of functions are updated to mining the structural hole spanners finally.Experimental results on real datasets and public datasets that validate the efficiency of our proposed methods are shown at the last of this dissertation. The experiments on co-author networks show that, it is reasonable to take community into account, in addition, W_HIS and W_HIS with different initialized importance of nodes are not influenced by community detection algorithms and the structural hole spanners are almost the same with each other.