Community Detection Methods Based On Density Peaks And Whale Optimization Algorithm

Network is a useful model-tool,which can be used to further study complex systems in the real world.Community is a special structural property in the network,which is a set consisting of closely connected nodes.The nodes in the same community have some commonalities.Therefore,community reflects the local characteristics of the network,and it can help people mine and analyze the deep knowledge contained in the network.Community detection is a main method which is used to detect communities in the network.However,network often has large-scale and complex structure,and the boundaries between communities are not obvious,which brings great challenges to community detection.In order to solve these problems,this thesis studies community detection from two aspects: overlapping and nonoverlapping.The density peaks algorithm and whale optimization algorithm are applied to community detection in this thesis.In addition,rough set theory is used to measure the similarity between nodes and divide overlapping nodes.The main research works of this thesis are summarized as follows.1.Aiming at the problem that the calculation results of local density and minimum distance are not ideal when the density peaks algorithm is applied to community detection,a topological structure-based density peaks algorithm for overlapping community detection(TSDP)is designed.TSDP calculates the local density by the degrees of node and its neighbors,and it can efficiently and accurately calculate the local density;TSDP uses cosine similarity to calculate the minimum distance,and a minimum distance discrete model is designed which increases the distinction of minimum distance and is conducive to selecting the center nodes more accurately.TSDP defines the core jump value to measure the change of the core value of each node,and the node with the largest core jump value is found as the boundary point,which improves the accuracy of center nodes selection.Experimental tests are conducted on artificial and real-world networks,and the experimental results show that TSPD is effective for dividing overlapping communities.2.Aiming at the low efficiency of density peaks in large-scale networks,an improved density peaks clustering based on rough set theory for overlapping community detection(RSDPCD)is proposed.RSDPCD calculates local density more efficiently and accurately according to the local aggregation coefficient,and all nodes in the network are divided into local peaks set and ordinary nodes set.RSDPCD defines different minimum distance calculation methods for different node sets,which effectively avoids the calculation of a large number of distances between nodes and improves the efficiency of the algorithm.The existing nodes' similarity measurement methods don't fully consider the nodes' connection relations,which leads to the result that the similarity results are rough and inaccurate.Rough sets are introduced to approximately describe the topological relations between nodes,and a rough subgraph model is defined to fully consider the nodes' connection relations.The model can measure the similarity between nodes more accurately.Aiming at the problem of inefficient and inaccurate division of overlapping nodes,the rough set is used to describe the community,and the overlapping parameters are continuously increased by iteration to continuously reduce and re-divide the upper approximation of each community in the network,and it can efficiently divide the overlapping nodes.The experimental results on artificial and real-world networks show that RSDPCD is effective.3.Aiming at the problems of limited search ability,easy to fall into local optimization,and high time complexity in the community detection algorithm-based swarm intelligence,a community detection algorithm based on whale optimization algorithm with evolutionary population(EP-WOCD)is proposed.EP-WOCD applies the strong and stable whale optimization algorithm to community optimization.Discrete improvements to the three search behaviors of the whale optimization algorithm make it more suitable for community detection.A variety of mutation strategies are added to enhance the diversity of whale populations,so as to improve the global search ability of EP-WOCD.The boundary node adjustment strategy is added to improve the local search ability of the algorithm,and the community consolidation strategy is added to eliminate the abnormally small communities in the results to further improve the quality of the results.In addition,in order to improve the efficiency of EP-WOCD,a multiswarm evolutionary population strategy has been designed.In the artificial and realworld networks,EP-WOCD is tested and compared.The experimental results show that EPWOCD is an effective non-overlapping community detection algorithm.