MutuRank Algorithm Based On The Research Of Heterogeneous Social Network Analysis

Community detection is a hot issue in the field of data mining research in recent years，withthe rapid development of network community，A large number of researchers have focused oncommunity detection research. The main content of community detection research is analysis onthe relationship between the existing in the network, and then to help people understand theorganizational structure of the network, it can also excavate a large number of potentialinformation in the community. In the real world, most Social Networks in the form of multiplerelations, so as the further research of community detection,“Heterogeneous Social Networks”concept was put forward, more and more scholars pay close attention to Heterogeneous socialnetwork analysis.“Heterogeneous Social Networks” is widespread in real life, such as Thescientific research cooperation network, social networks, medical health network, etc. Along withthe coming of the era of big data, how to deal with the large-scale complex heterogeneousnetwork structure has been a challenging problem.At present, There have been many studies in the literature targeting at discoveringcommunities from social networks. However, most of them have focused on single-relationalnetworks. A hint of methods detected communities from multi-relational networks by convertingthem to single-relational networks frstly. Nevertheless, they commonly assumed differentrelations were independent from each other, which is obviously unreal to real-life cases Thecurrent study methods are considered that the relationships between different heterogeneousnetworks are independent, and each relations are equal. However, they are not reasonable in reallife. In this paper, we attempt to address this challenge by introducing a novel co-rankingframework, named MutuRank, this is the first innovation point of this paper. It makes full use ofthe mutual infuence between relations and actors to transform the multi-relational network to thesingle-relational network. At the aspect of spectral clustering, we then present GMM-NK(Gaussian Mixture Model with Neighbor Knowledge) based on local consistency principle toenhance the performance of spectral clustering process in detecting communities, this is thesecond innovation point of this paper.In this paper, the main work as follows：(1) This paper first introduces the research status and research significance of communitydetection, and the advantages and disadvantages of the existing community detectionalgorithm has carried on the detailed analysis and comparison.(2) In allusion to “Heterogeneous Social Networks”, in this paper, we attempt to put forward a novel co-ranking framework, named “MutuRank”, and assuming different relationshipbetween is not independent, using the mutual influence between relationship and nodes toiterative calculate the weight of relationship, and get the equilibrium distribution ofrelationship. At last, the multi-relational network is transformed to the single-relationalnetwork.(3) In terms of community detection, GMM-NK(Gaussian Mixture Model with NeighborKnowledge) are put forward in this chapter. Doing spectral clustering analysis Using GMM-NK algorithm on synthesis networks that obtained from“MutuRank”, Compared to thetraditional GMM algorithm, GMM-NK algorithm combines the Neighbor Knowledge.Experimental results both on synthetic networks and the real-world network have verifed theeffectiveness of MutuRank and GMM-NK.(4) Finally, two “Heterogeneous Social Networks” are constructed in experiments by using twoUCI data sets Iris and Breast, and extract the DBLP data set as the real-world dataset to dosome experiments. Through MutuRank experiment, the equilibrium distribution ofrelationship would be botainand, do spectral clustering through GMM-NK to validate thethe validity of the experimental results eventually. In terms of performance appraisal, we useNMI (normalized mutual information) as evaluation indexes. A series of the experimentalresults show that, no matter in the synthesis of heterogeneous network or in real socialnetwork, MutuRank algorithm and GMM-NK algorithm is effective.As for theoretical and experimental, it can be proved that MutuRank algorithm andGMM-NK algorithm can be used in the real world to the detection of heterogeneous socialnetworks, and the effect is remarkable. With the advent of the era of big data, social networkanalysis will be even more important, social network analysis is meaningful and significant tonational security monitoring and business, this field is worth the long-term in-depth explorationand research in the future.