Research On Community Detection Algorithm Based On Hierarchical Clustering

There are various systems with a specific function in the realistic society,such as electronic commerce system,scientific work system,online social system,these systems can be abstracted as a network with complex internal structure,called complex network.Many studies have shown that complex networks have a very important feature-community structure.Community structure can be understood as a collection of similar elements in the network and it is of great significance to understand the structure and function of complex networks.Community structure is often hierarchical,such as the regional relation network,if we use resident cluster as the standard to detect community,then these communities are nested,because the village is included in the town,the town is included in the county,the county is included in the city and so on.In order to detect the hierarchical structure of complex network,this paper studies the problem of community detection in complex networks from the perspective of hierarchical clustering,including the following contents:(1)We propose a hierarchical community detection algorithm based on similarity.The traditional Louvain algorithm is extremely fast,but the accuracy of detecting communities needs to be improved.This is because modularity of Louvain only considers link information between nodes and neglects the effect of the surrounding neighbor nodes,leading to decreased tightness between nodes in the same community and affecting accuracy.To solve this problem,by introducing node similarity to improve modularity function of Louvain algorithm,we propose a hierarchical community detection algorithm based on similarity.Experiments on real network and LFR artificial neural network show that our algorithm can improve the accuracy of hierarchical community detection.(2)We propose a hierarchical overlapping community detection algorithm based on maximum clique.Study shows that many real networks have overlap structure,in order to detect hierarchical structure and overlap structure,we use maximum clique instead of node in traditional hierarchical clustering algorithm and define the maximum clique extension strategy,then we propose a hierarchical overlapping community detection algorithm based on maximum clique.The algorithm uses the maximum clique extension strategy to generate a hierarchical dendrogram,and then use the overlap modularity function to select the optimal results.Experiments on real network and LFR artificial neural network show that the proposed algorithm can achieve a good level of community and accurately identify the overlapping nodes.