Research On Key Node Mining And Recommending Algorithms Of Social Network Based On Graph Theory

Graph Theory takes graph as research object,which is the most commonly used modeling language and analysis tool in studying various real networks.Using Graph Theory in network studies will reduce the complexity of those researches.Social networks are extension of human social activities.Similar to the real world,the core figures of social networks will still play an important role in the local radiation of information based on their prestige,trust,and activity.Based on positional relations,the connectivity key people between social network regions will have an inestimable power in information penetration among networks.The paper focuses on the selection of the main node and the type of recommendation information in the social network information dissemination.The research includes the mining of these two types of key nodes in social network and information recommendation technology.The purpose is to improve the guidance and control of public opinion,and to provide users a more efficient information service.This paper mainly includes following researches and results:(1)A key node mining method based on hierarchical filtering is proposed,which can reduce the scale of mining algorithm while ensuring the quality of mining.The precondition of the hierarchical filtering is replacing a single indicator with hierarchical indicators system.The evaluations of the central node in the paper are degrees and clustering coefficients.The combination of degree and clustering coefficient can not only filter the calculation amount of node,but also ensure a larger range of information penetration ability than a commonly used single index clustering coefficient.The evaluations of connectivity nodes in the paper are embedability and betweenness.The embedability is also a simple locality index.The betweenness is an index for mining high precision,but needs global calculation.Because of the consistency of the embedability and the betweenness,the embedability can be used as the pre-filtration before the calculation of the betweenness.(2)The design of an improved betweenness calculation method based on flow rate.The calculation scale is apparently reduced by pre-filtration and the limited-layer calculation.Improvements include flitting candidate nodes with the embedability indicators,considering the variations in degrees of traffic from different nodes,and taking finite layers in the broad-first search.The effect of filtering is to eliminate the sub-trees that do not need to be calculated in the process,which can greatly reduce the computational scale;The design idea of finite layer computing comes from the inhabitation of the information spread that caused by the category attributes owned by both information and user nodes.Corresponds to a broad-first search,the larger the gap between levels is,the greater the difference of categories is.In the aspect of information transmission,a greater difference means less information can travel from the top layer to the lower layer.Thus the lower-level flow can be appropriately ignored,in order to reduce amount of calculation.(3)The design of social network recommendation algorithms based on selective heat conduction/substance diffusion,which fully exploits the respective advantages of the two algorithms and reflects the mutual influence among social network users.The paper firstly demonstrates the features of the two recommendation algorithms by process demonstration methods,which is a supplement to the current experimental verification.It points out that these two types of algorithms are still similar in nature to the recommendation algorithms based on collaborative filtering;Based on the strong influence of key nodes on information dissemination,the research of the social network recommendation algorithm in this paper is based on the recommendation of the key nodes in the network.The research of the social network recommendation algorithm in this paper is based on the premise of recommending to the key nodes in the network.By using the distinctive information dissemination function of the two types of key nodes,the recommendation algorithm based on substance diffusion with higher accuracy is recommended for the central node.In the algorithm,the recommended heat transfer-based recommendation algorithm with good diversity is the connectivity node recommendation algorithm to maximize their respective advantages;At the same time,considering the mutual influence of nodes in the social network,the paper improves the two types of algorithms based on the centrality of eigenvectors.(4)The algorithm proposed in the paper is verified experimentally by Spark,the parallel computing framework that is particularly suitable for graph operation.Experimental design includes evaluation method design,acquisition of experimental data sets,optimization of experimental platform performance,and analysis of experimental results.The number of people affected in the SIR model is the evaluation index of the key node mining algorithm.The experimental results show that the key node mining algorithm proposed by the paper has a good performance.The proposed algorithm is evaluated by accuracy and diversity.The experimental results show that the social network recommendation algorithm based on selective heat conduction/substance diffusion that proposed in this paper improves the diversity of recommendation results while ensuring certain accuracy.