Research On Centralization Algorithm Of Large - Scale Network Analysis

With the rapid development of the Internet in recent decades, the enormous increase in the number of users cause the user data is constantly mass produced, which are hundreds of millions or even tens of millions of ultra-large data sets. So, how to get deeper useful information, excavate business value, understand business practices and find new business growth from these vast amounts of user data has become an important research directions and challenges. But the rapidly expansion of the Internet has led to a rapidly increase in the size of the user data size, bringing about new demands and challenges dealing with classic network analysis algorithms and tools, which names the ability of processing and analyzing large data sets. This article aims to solve large-scale network analysis problem on multi-core PC(personal computer).Centrality analysis is the focus of the social network analysis, what the powers of the individual or organization possessed in social network, what the central positions of they located, is vital to the information spread in the entire network and the effective of information spread.With the development of computer science and parallelism algorithms, we choose parallelized centrality algorithm to solve the large-scale network analysis problem on PC. Following is the main achievement by this article:1.We will inspired by the feature of small world, integrate the principle of divide and rule, design parallel algorithm on degree centrality and closeness centrality respectively, especially on betweenness centrality algorithm, take the graph traversal process as a independent running module, assign it computational processor, parallelize it, propose fine-grained and coarse-grained parallel algorithms based on betweenness centrality algorithm.2. This article will give full consideration to the characteristics of the unbalanced degree distribute, removing a large number of vertex that degree is 1 in real world, revising the algorithm code, so as to improve and optimize the parallel algorithm.3. This paper will choose MATLAB parallel toolbox, using the data parallel programming named ‘parfor’ to implement the betweenness centrality parallel algorithm.4. Finally, on the same platform of multi-core PC and MATLAB, this paper achieve different experiments with parallelism and non-parallelism algorithms on various scale data sets, so as to test and verify the performance of algorithm parallelism. We take an application analysis of betweenness centrality parallel algorithm in Yeast’s protein-protein interaction networks, then discover an associated information: protein that have high degree centrality value more likely obtain the high betweenness centrality value.