Study On The Key Problems Of Information Dissemination In Social Networks

With the rapid development of network technology and the prevalence of the Internet,social network applications such as We Chat,Weibo,Tik Tok,and Twitter have been infiltrated into all aspects of people's daily life.These apps play an important role for people to obtain,communicate and disseminate hot social news and thus become an indispensable part of people's life.Information dissemination on social networks owns many characteristics,such as fast in speed,wide in coverage,and prompt in time.Some hot news would quickly brew into a strong public opinion in a short time,which could impact the evolution of social affairs to a certain extent.However,due to some social information is partial and misleading,they would inevitably result in some superficial and even incorrect public opinions,which will have a negative impact on social stability and even damage national security.Therefore,analyzing and understanding the behavior of information dissemination in social networks,revealing the characteristics and laws of information dissemination,identifying the key nodes in the dissemination process,and inferring the hidden diffusion network structures,will help to guide the evolution of public opinions,alleviate social contradictions,reduce the negative impact of unexpected affairs,and improve the credibility of the government,so as to promote the healthy and harmonious development of the society.To this end,this paper studies the key problems of information dissemination in social networks,which mainly includes four aspects as follows.(1)To solve the problems that the numerical solutions of current information dissemination model equations do not match the simulation results of node behavior,this paper proposed a novel discrete social network information dissemination model,i.e.SPIR(Susceptible-Potential-Infective-Remove)based on the potential spreaders.Potential spreaders refer to the susceptible node which has a direct connection with infective nodes.The potential spreaders could optimize state variations of the information dissemination process and avoid the repeated calculation problem in model equations.By introducing the potential spreaders and analyzing their complex variations during the dissemination process,this paper constructs the discrete dissemination equations which extend the classic SIR model.Experiments on various datasets and under different parameters show that the proposed SPIR model could accurately depict nodes' state changes in the real information dissemination process and maintain a high degree of fitness with the simulation results.(2)When evaluating nodes' centrality,existing influence maximization algorithms are low in resolution and most of them are based on local measurements.Thus,this paper proposes a novel influence maximization algorithm for social networks,i.e.RNR(Reversed Node Ranking)based on adaptive node reverse ranking.The proposed algorithm exploits the reverse order which could reflect a node's importance as the weight to evaluate a node's influence.Based on the measurement,the algorithm sorts the nodes iteratively until convergence,which guarantees the high resolution of nodes' evaluation.Besides,to avoid the rich-club effect,this paper proposes two seed selecting strategies.One ensures that there is no connection between either of the seeds and the other chooses to weaken the seeds' neighbors.These two strategies are suitable for Independent Cascade Model and Weighted Cascade Model respectively.The abundant experiments on the two dissemination models show that the proposed RNR algorithm has excellent performance.It also reveals that the sensitivity of Independent Cascade Model and Weighted Cascade Model towards the rich-club effect is significantly different.(3)For the problem that the existing network dismantling algorithms ignores the importance of the tree-breaking step and misses the data pre-processing of networks,this paper proposes a novel social network dismantling algorithm,i.e.SEGTB(Skeleton Extraction and Greedy Tree Breaking)based on the network skeleton and a greedy strategy.The network skeleton is the connected subgraph comprising of a small group of nodes that supports the topology structure.The skeleton covers as many as possible neighbors and meanwhile has fewer connections among themselves.The proposed algorithm first extracts the network skeleton,which serves as an effective data pre-processing step.Then,the algorithm decycles the remaining network by iteratively pruning 1-degree nodes to obtain the acyclic graph.At last,the acyclic graph is dismantled through a greedy tree breaking method which guarantees to remove nodes whose amount is strictly less than twice of the optimal solution theoretically.Extensive experiments on ten real-world networks show that the proposed SEGTB algorithm has a better performance compared with other baselines and always keeps high efficiency.(4)When inferring the structure of social diffusion networks,existing algorithms are usually based on a large number of cascades and only take into account the time difference between two nodes to measure the edge probability between them.Thus,this paper proposes a novel diffusion network inference algorithm,i.e.NPDE(Normalized Probability and Degree Estimation)based on the normalized edge probability and the estimation of nodes' degrees.The proposed algorithm first calculates the independent edge probability within each piece of cascade according to specific probability distribution models.Then,all the edge probabilities that point to the same node are normalized based on their independent probabilities.After handling all cascades,the proposed algorithm evaluates the edge likelihood between two nodes by summing all normalized edge probabilities between them.Due to the normalization,the proposed algorithm could maintain a good performance with a limited number of cascades.Besides,the algorithm estimates the nodes' degree based on their corresponding edge likelihoods and finally infer the network structure under the restriction of nodes' degree.The experiments on four network generating models and three probability distribution models show that the proposed NPDE algorithm keeps a good performance as well as a high efficiency with a limited number of cascades.It also reveals that the restriction on degree brings a significant improvement for inferring the diffusion networks.