Research Of Key Techniques Of Community Detection And Search On Complex Networks

With the rapid development of information technology,complex networks from sociology,physics,economics,biology,computer science and other fields have shown more diverse and complex structural features.Community is an important feature of complex networks.Nowadays,community detection and community search are two main techniques for analyzing and mining communities on complex networks.Community detection aims to find all communities in complex networks,and community search aims to find the local communities that contain the given nodes.The main difference between them is that community detection focuses on the whole complex networks and uses global standards to retrieve and analyze all of the communities integrally,while community search starts from one node or several nodes,and uses personal evaluation methods to search and analyze the characteristics of the local communities.With the development of deep learning and big data,community detection and community search on complex networks are of great significance in exploring the network structure,and have been paid more and more attention by academia and industry.1.This paper proposes a community detection method based on deep sparse autoencoder.The high-dimensional adjacency matrix of complex network cannot reflect the local information of each node integrally and cannot express the main features of the network topology.Thus,it has weak expression to the community structure in the network,which affects the accuracy of community detection.This paper attempts to use a deep learning method to solve the community detection problem.First,a hop-based matrix processing method is proposed to optimize the sparse adjacency matrix and obtain the similarity matrix.Then,a deep sparse autoencoder is constructed to extract the features of similarity matrix and get a low-dimensional matrix.Finally,we use k-means to compute the communities.Experimental results show that this algorithm can obtain more accurate communities than the baselines.2.This paper proposes a community search method based on attributes.Most state-of-the-art community search methods only consider the network topology,but ignore the effect of node attributes which contain abundant information.This may lead to the inaccuracy of the predicted communities.This paper attempts to combine the information of network topology and node attributes to solve community search problem.First,the concepts of topology-based similarity and attribute-based similarity are presented to construct a TA-graph.The TA-graph can reflect both the similarities between nodes from the respect of the network topology and that of the node attributes.Then,AttrTCP-Index is constructed based on the structure of TA-graph.Finally,by querying the AttrTCP-Index,the communities are found out for the query nodes.Experimental results on real-world networks demonstrate that this algorithm is an effective and efficient community search method.