Research On Adaptive Clustering Algorithm Based On Data Field In Complex Networks

Complex network is a new subject which is developing with other researchs, and it has obvious cluster characteristics of network topology structure. The network cluster structure is one of the most important and basic attribute like the small world and scale-free characteristics of complex network that is the verters in the same cluster are connected closely with each other while interconected sparsly in different clusters. The porpose of the clustering algorithm in complex networks is to discover the real and exists clusters in complex network, and researching on clustering algorithm has very important pratical value which can help us to discovery the cluster structure in complex network so that we can understand the functions of it, for example, predicting the development of the complex networks, analysising the behavior of the social networks, finding the hot topics in web networks and so on. Thereofore, this disseration mianly studies on the clustering algorithm which is to detect the cluster structure. Follow is the main work of this paper.Firstly, this paper presents the background, pratical significance, and then mianly states the research status of complex network and the clustering algorithm, also focuses on the research situation at home and abroad of the clustering algorithm in complex networks, manily researchs on the node centrality and the clustering objectice function, elaborations two kinds of classical clutering algorithms.Secondly, through the research and ananlysis on the detection of cluster structure of clustering algorithm in complex networks and data clustering algorithm, we can find that the two kind algorithms are very similar. And digging out the cluster center nodes can grate help to reduce the complexity of cluster algorithm, while the research on evaluation of the node importance is suitable for digging those nodes. By analyzing and comparing the advantages and disadvantiges of various node importance evaluation algorithms, this paper presents the concept of node importance factor to detecte the cluster center nodes in network. After study on the merits and shortcomings of the existing method which is used to find cluster in complex networks, we find that the clustering evaluation function can solve the difficult problem in initial cluster number by constructing a suitable clustering evaluation function. According to the above ideas, this paper combines the idea of k-mean algorithm and the character of Gaussian function to porpose an adaptive method to detect clusters which is based on data field in complex network.Thirdly, considering if only make the important node as the cluster center will be not easy to find the big difference of the cluster structures in network, especially unsuited for the core and periphery structure network, and it exists one-sideness on a single clustering evaluation function. With the purpose of overcoming the defects of the last method, this dissertation will bring in the vertex similarity to avoid all important vertices becoming the cluster center vertex. At the same time, it introduces the multi-objective clustering evaluation function to detecte cluster structure better and more effective. According to this idea, this paper porposes another algorithm namely a multi-objective adative algorithm based on data field.Finally, simulation results show that these two kinds of clustering algorithm can effectively find cluster structures and the time complexity is lower than the general clustering algorithm.