Research On Overlapping Community Detection Based On High Leadership Seeds Extension For Directed Networks

With the innovation and development of information technology,complex networks have become increasingly crucial for designing and optimizing complex systems and profoundly exploring the evolution of network structures.As a fundamental feature of complex networks,community structure can explore networks’ potential topology and dynamic evolution laws.Therefore,many community detection algorithms have emerged in recent years.Among them,the more typical is the local extension method,which can discover the overlapping community structure with only partial network information,but there are problems such as uneven seed quality and low efficiency of the algorithm;in particular,the direction and weight information of edges in the real network is of great value to exploit,while most algorithms are limited to the analysis of community structure of undirected and unweighted networks,ignoring the latent relational structure of edge information.Therefore,combining the shortcomings of local extension methods,this paper introduces the direction and weight information of edges and reasonably defines process indicators and evaluation criteria to propose a directed network overlapping community discovery algorithm based on high-leadership seed extension.The main research contents are as follows.(1)To solve the shortcomings of local expansion methods and insufficient consideration of directional information,this paper proposes a directed network overlap community discovery algorithm based on high-leadership seed expansion(LSEDC).Firstly,this paper integrates the local and global information of nodes,incorporates the directionality of connected edges,and selects the seeds with high leadership and information transmission ability as the starting point of community expansion.Second,a seed community formation strategy based on the principle of the k-order neighborhood following degree and density maximization is proposed to mine the community core precisely.Subsequently,a community neighbor layer expansion approach is adopted for node aggregation,and the affiliation with the community is corrected by node movement driven by conductivity.Finally,community reoptimization is performed by assigning free nodes and merging highly overlapping communities to obtain the final division results.Through comparison experiments on the actual network and the LFR benchmark network,it is verified that the proposed algorithm can effectively improve the seed quality and distribution and improve the accuracy and stability of the overlapping community discovery results of the directed network.(2)To improve the efficiency of directed network community discovery and extend it to weighted networks,this paper further considers the connected edge weights.It introduces a Hadoop-based Map Reduce computing framework to propose a parallelized community discovery algorithm for directed weighted networks based on highleadership seeds(P-DWLS).The method can accurately measure the leadership of nodes in a directed weighted network,screen the maximum weight points as candidate seed nodes,form seed communities and neighbor layer extensions in parallel style,correct node positions and optimize communities with weighted derivatives and weighted affiliations as a guide.Experiments show that the P-DWLS algorithm can discover high-quality community structures on the directed weighted network with high generality and flexibility while improving the algorithm’s efficiency,thus verifying the rationality and effectiveness of the research work.