Study On Overlapping Community Detection Via Local Optimization

There are many things in the real world can exist in form of networks.The online social,the relationships between people,the models of protein structure inside our body and so on.As we make further study into the complex networks,the community structure is becoming more and more important as unique feature in the networks.The community structure can reveal the relationships between different nodes,the study of community structure in the complex networks can help people to get deeper understanding of the structure on network system and its evolutionary pattern.When we apply this on improving our real life,it will be very useful.This is the very reason why the study of community detection becomes important in the area of complex network research.The detection of overlapping community can reveal the inner relationships in the network more precisely,and it had brought more and more attention in recent years.Speaking of overlapping community detection,there are a lot of research results been brought out.The main methods of overlapping community detection can be divided into global optimization method and local optimization method.The former methods carry the community detection from a global perspective,they need to get the overall information of the whole network,including graphic partition method,hierarchical clustering method,modularity optimization method,spectral clustering method,method based on model and so on.The local optimization method always start the community detection with a few nodes,such as expand local optimization method,label propagation method,fraction filter method,local optimization method and so on.In this paper,our main work is based on LFM algorithm.Considering the algorithm itself have many weaknesses,such as high time consumption and community detection result unstable.In order to solve these problems,we made some improvement.To solve the high time consumption problem,we proposed the definition of ‘core area' and proved that when the node belongs to the core area of one community,we do not need to calculate the value of its fitness.We also redesign the strategy of expansion on community.In order to improve the stability of the community detection result,we bring the threshold to limit the node being seed nodes.Comparing with GCE,our algorithm will consume less time.In the end,we test our improved algorithm on real networks and LFR networks,with Qov and NMI separately to evaluate the quality of our improved algorithm.And the result reveal that our improved algorithm can generate better result than the origin one and CPM,Copra.