Studies On Influence Maximization Based On Node Classification In Social Networks

Influence maximization,as one of the relevant research contents in the field of social network analysis,aims to find the most influential seed set in the network as soon as possible,and maximize the influence scope through the interaction between the seed nodes and other nodes.With the advent of the era of big data,the scale of the network grows rapidly,and it is urgent to study how to effectively solve the problem of influence maximization within large-scale social networks.Since real social networks often have obvious community structure,and searching for influence nodes in each community can effectively reduce the computing cost.To this end,this thesis uses the node classification algorithm to divide the large-scale social network into multiple communities,and then studies the influence maximization on this basis.The specific contents of the study are as follows:1.Aiming at the problem that most existing node classification algorithms based on network embedding ignore the complementary between network structure and attribute information,this thesis utilizes the network representation learning method to learn the lowdimensional representation vector of nodes and unstructured attribute information is represented as low-dimensional information features that are easily processed by the computer.Furthermore,a multimodality fusion model is introduced to integrate knowledge from the network structure and attribute information by imposing consistency constraints on the prediction labels of different modalities in the feature space,so as to further improve the accuracy of node classification.Finally,a multilayer neural network based on softmax regression is used to predict labels.Experimental results on several public datasets demonstrate the effectiveness and robustness of the proposed algorithm.2.Aiming at the problem that the influence maximization algorithm based on global network is often difficult to be applied to large-scale social network due to its high time complexity,this thesis combines the community structure and divides the solution of influence maximization into the candidate stage and the greedy stage.In the candidate stage,heuristic algorithms are used to select the highly influential nodes from the internal and boundary of each community to form the candidate node set.In the greedy stage,the submodular property based greedy algorithm is utilized to select the seed nodes with the largest marginal influence increment from the set of candidate nodes,where the influence propagation scope of the candidate nodes is restricted to the communities where the nodes and their neighbors are located.Experimental results show that the proposed algorithm significantly reduces the running time while ensuring the influence range.