Research On Overlapping Community Detection In Weibo Networks

With the rapid development of Internet and mobile communications technology, more and more netizens have used social system to exchange and share information, which promotes forming an increasingly large social network. This logical network built upon physical network could reflect users’preference and social relationships between them. How to discover the characteristics and potential value of the social network has become a problem which has attracted great attention. As one of the main networks (especially social networks) research, community detection is of great significance to research on network topological characteristics, so this thesis chosen weibo network as the object, studied on the network with overlapping structure, and proposed an overlapping community detection method.The main work of this thesis includes the following aspects:● This thesis discussed the development and state of art of community detection from three aspects. From the theory of complex networks standpoint, this thesis analysis and discus the principles, characteristics, advantages and disadvantages of traditional community detection algorithms; analyzed and summarized the unique characteristics of weibo networks, and summarized the improved community detection algorithms in weibo networks; summarized key technologies of community detection based on genetic algorithm.● This thesis proposed an overlapping community detection method based on GA for weighting weibo network (WOGA), and this algorithm is divided into two sections, weibo network weighting method (WNWM) and overlapping community detection based on GA (OCDGA). WNWM algorithm considered static topology, interaction frequency, and similarity of topics and labels of users, and base on this WNWM built a user relationship strength evaluation model, and WNWM weighted the edges between weibo users. OCDGA improved the existing matrix coding scheme to encode the individuals. And OCDGA improved the module function of overlapping communities by using weighting networks and use this function to calculate individual fitness. OCDGA proposed a population initialization approach based on principle of node centrality and similarity, and designed crossover operator and mutation operator which could be suitable for overlapping community structure and matrix encoding method, and designed selection operator based on elitist strategy. Lastly, OCDGA introduced a self-adapting migration strategy to improve the accuracy and efficiency of algorithm.● This thesis gave a distributed implementation scheme of WNWM and OCDGA based on MapReduce. In the process of the distributed implementation of OCDGA algorithm, the double-coarse-grained PGA model is proposed to design the details of population migration rules.This thesis has done experiments on random network, classical real network and weibo networks to validate the accuracy of WOGA, and compared the experimental results of WOGA and traditional community detection algorithms. Compared with other algorithms, the experimental results proved that WOGA can get more accurate result especially in weibo network.