Research On Overlapping Module Discovery And Noise Processing In Complex Networks

Complex network is an abstraction of complex systems. It can be seen everywhere in our daily life, such as social networks, protein interaction networks, disease transmission networks etc. In the Internet era with the rapid development of information technology, the study of complex networks has become one of the new research focus. Community structure is a unique and real structure based on complex networks, which are mapped to real complex network of different functions or different structural units. It plays an important role in analyzing of the topological structure of complex networks, understanding its functions and searching for its potential property.Currently, although there are some algorithms such as CPM, LINK algorithm can be found good community structure in a certain extent, but complex networks with large amount of data, complex structure, the characteristics of noise. Therefore, in the discovery of overlapping community, noise processing, accuracy, speed and so on also need to a lot of research work. It is found that there are a lot of methods in the society, such as data mining,the method of matrix decomposition and so on. Clustering is the most commonly used method of the relationship, as well, this paper will be applied to the idea of clustering complex network of community structure.Overlapping community discovery and noise processing is very complex. Existing algorithms exist excessive overlap, weak reliability data, for excessive overlap problem, this paper presents found overlapping communities linkw algorithm based on dynamic characteristics of complex networks is proposed for noise processing of MG algorithm. The main contributions are as follows:1. Overlapping community detection. Existing overlapping community detection algorithm, most of the them have the problems of excessive overlap, unreasonable division and so on, an overlapping community discovery algorithm based on weighted edge similarity is proposed for the problem of excessive overlapping(LINKw). Firstly, we transformation of the original network into the corresponding edge graph, and then using the cosine similarity method of its similarity calculation sequentially combined similarity large node(ie, the network edges in the graph), in order to find overlapping community structure, A qualityfunction of community structure is proposed, which is dependent on the interaction coefficient and the density. Our experimental results demonstrate that the algorithm performs better than some popular methods.2. The noise processing. Protein interaction network is a typical complex network.Protein interaction network in the presence of large amounts of noise data, existing algorithms widespread costly, time-consuming and more features. According to the dynamic characteristics of complex networks, the noise problem is one of the hotspots in current research. A framework model based on loose in and strict out is constructed in dynamic protein interaction network, and we do it by fusing the module function and the gene ontology.Currently, it is very popular in dynamic protein interaction network to filtering noise. We find that the proposed method can get better results.