Study On Effective Community Detection Algorithms Based On Structure

With the development of science, people pay more and more attention to complex network analysis. Complex network analysis becomes a research hot topic in the community of machine learning and data mining. In complex network, a typical is social network where the members are contacted and interacted each other. The system under social network is stable and has some interesting properties. In social network, the community refers to the group in which members linked closely. Generally, the members in one community have the same characteristics in some aspects and internal individuals in the same community are susceptible to others. Effective community mining technology can discover and display the real community structure. Analyzing large-scale network is important and practical significant with the development of social computing. However, the traditional techniques applicable to large-scale complex networks are hard to meet the requirements on both computational complexity and identification accuracy.In this thesis, we introduce the structural similarity information and fitness into the community detection process. Nodes which share enough common neighbors should be to large extent grouped into the same community and the final detection accuracy could be improved by adjusting the fitness value. This strategy makes the algorithm effective and efficient in term of running time and modularity. Moreover, we apply the structural similarity information in ant colony optimization model for fastening the learning process. The method has ability to consider both local information and global characteristics of complex networks because of addition of the structural information. It also reduces the computational complexity by making use of the swarm intelligence. Experimental results on synthetic complex networks and real social networks have shown the outperformance of the proposed methods by comparing with the typical existing methods.