Research On Graph Data Mining Technology In Social Network Analysis

With the development of information processing technology and Internet technology,Internet visits has gradually increased and formed large-scale virtual networking data.In a large and complex networking data set,it is very important to mine valuable information,especially the analysis of community discovery and related problems have been regarded as a research hotspot.However,there are still many challenges in the field of graph mining: firstly,the size of topology graph is becoming larger and larger in practical applications,and traditional graph mining methods are only suitable for small or medium scale data;secondly,when the topology is more large,more complex,many edges will carry redundant and false information,and this kind of noise is easily to influence the inherent structural properties of the graph.This situation causes two results: one is to induce futile computing;the second is to produce low-quality of graph clustering.These challenges make the traditional graph clustering low efficiency and precision;the great increasing number of edges and points cause low efficiency of programs and relatively low utilization rate of hardware resources.Considering the above challenges,this paper has introduced the background and significant,and analyzed researching status at home and abroad,the typical graph clustering method.In order to adapt to the complex structure and big data of complex network data,this paper has researched the related works,and proposed a graph mining model using Markov clustering based on annular network motifs(gmm Mcanm).The major innovations include:1.A determination method is proposed based on annular network mode.Firstly,a set of random graphs are generated by Erd?s-Rényi model according to the point and edge set of input graph.Then,proof that the additive property can be used as a judging conditions for the annular network subgraph.Finally,we construct two quaternary structures.In the process of subgraph mining of input graphs and stochastic graphs,we calculate two statistical features of ring subgraphs: P_value and Z_score to determine whether the subgraph is a motif.The method uses a simple data structure,and the statistical characteristics of the graph are accurate and fast.2.A graph clustering model based on motifs are constructed.Firstly,the contribution of each edge is quantified,and we solve the correlation matrix of the absolute contribution of edges in topological graph.Then,the matrix is binarized by the contribution threshold obtained by the dynamic threshold method Ostu.Finally,a flow process is simulated: after the elements of the self-return and each node in all columns is normalized,the Markov matrix is formed.The matrix is subjected to iterative expansion and expansion to achieve the convergence state.NMI and F-score are used to evaluate the clustering results.Experimental results show that the proposed model can effectively reduce the running time and improve the mining efficiency of the graph when the clustering quality is guaranteed.In the field of data mining and artificial intelligence research,graph mining is regarded as an important research direction,and it is most active and most effective method in the study of large and complex network.The major research work in future is to focus the role of network motifs in the different types in graph mining and to find the best stochastic graph construction model.