An Community Detection Algorithm Based On Attribute-relationship

With the rapid development of the Internet and technology, people through onlinesocial networks are becoming more and more close, especially in recent years,network development, social information through QQ, micro letters, microbloggingand other close around the distance between people, people with common interestsand personality are linked to the social network community is becoming more andmore obviously, in order to make a better evolution of social network to better servethe people, a large number of researchers begin to study community structurediscovery. Study on community structure should be made of all internal nodes of eachcommunity with each other as much as possible the node between the differentcommunities of the relationship is relatively distant, and the similarity between nodesto be larger within each community, based on these characteristics, in order to makethe quality of the community structure partition is better, the effective use of topologythe structure information of node attribute information in the network and the networkis particularly important for community structure discovery.Most of the current community discovery algorithm is split hierarchy clusteringtheory and graph based on the idea of these methods, although in most cases can be agood way to identify the community structure, but they have a common problem, thecommunity detection algorithms are mostly based on the information of networktopology is divided, and the time complexity is high, it is not scientific in real life,because everyone has their own characteristics and properties, only depends on thetopology information is single, and can not fully explain the relationship between allnodes within a community closely and similarity. In order to solve this problem, thispaper found that the correlation algorithm in the research community, the topologicalinformation of all nodes in the network have their own attributes of information and network, proposed algorithm found a use node attribute information and therelationship information comprehensive similarity effective community, which hasbeen tested and improved. The algorithm finds the algorithm is different from manyof the topology information of cluster community, it uses the node attributesinformation in real life to get the similarity between each other, and then combinedwith the relationship between these nodes to obtain the comprehensive similaritybetween each other, which combines k-means algorithm and on the basis of thecombination of unique the advantages of SVM are optimized. The experimentalresults confirmed the relationship between all nodes within the community because ofthe joining node attribute information becomes more and more closely, the quality ofthe community structure is relatively high.According to the analysis above, the main work of this paper are as follows:First, the attributes of nodes and the network topology of the network ofsimilarity extraction and fusion, and then according to the idea of K-means presents adetection algorithm based on KARIS comprehensive similarity degree of community,and has carried on the detailed introduction.Then, aiming at the shortage of KARIS algorithm in the algorithm of supportvector machine was introduced, and the center of a single node is solved by theintegrated similarity comparison by the community as a whole the disadvantages, andgive a detailed introduction of the improved algorithm.Finally, in the micro blog data contains the node attributes and topologicalstructure of the set of two methods before and after improvement is verified, theexperimental results show that the superiority of the improved KARIS algorithm andthe improved KARIS algorithm, at the same time for the tested in containing onlytopological structure of the data set, the experimental results show that the algorithmis also applicable to non-node attribute information network, proves that the algorithmis universal.