Research On Overlapping Community Detection Algorithm Based On Label Influence Value

With the continuous development of information technology,the complex network community structure has been more and more concerned by the academic community.The related technologies discovered by the community have become the focus of the current analysis of the online community.Therefore,the in-depth theoretical research on community discovery has extremely far-reaching significance and purpose for the community network structure and community network characteristics.At present,the division of community detection algorithms is mainly divided into two aspects,one is a non-overlapping community detection algorithm,and the other is an overlapping community detection algorithm.However,the reality is not as simple as imagined.It is a complex network structure.That is,a single node in a complex network may exist in multiple community network environments.Therefore,the research on overlapping community detection algorithms in this paper has great significance both in theoretical basis and in reality.The traditional tag propagation community discovery algorithm is the LPA algorithm,which uses the network structure alone to guide its process,neither requiring any parameters nor optimizing the objective function.It starts with a configuration where each node has a different label.In each step,a node(asynchronous version)or each node(synchronous version)decides to change its label to the label carried by the largest number of neighbors.If there are multiple tags,one of them is randomly selected as a tag.By building algorithm functions,each node has more neighbors in its own community than neighbors in other communities as it continues to iterate and functions gradually converge.More importantly,it is the basic algorithm for overlapping community discovery algorithms SLPA and COPRA.Both algorithms are based on this community discovery algorithm.The current common overlapping community detection algorithms include faction filtering CPM algorithm,SLPA and COPRA.Both SLPA and COPRA are new algorithms based on label propagation algorithm LPA.Compared with CPM algorithm,they both have better performance and complexity,and they are widely used and researched in the field of community detection.The COPRA algorithm can effectively mine overlapping community networks,but the algorithm has strong randomness,poor robustness,and the result of community division is extremely unstable.Aiming at the problems brought by the above-mentioned community detection algorithm COPRA,this paper proposes an overlapping community detection algorithm based on label influence value.The main idea is still to use the COPRA algorithm as the relevant foundation.Based on this algorithm,the concept of label influence value is proposed.Due to the low efficiency and poor stability caused by the COPRA algorithm in the label initialization phase,the randomness brought by the label selection phase is strong,this paper first proposes a label initialization method that does not overlap triangles for the problems of low efficiency and poor stability.This method ensures that the time complexity of updating labels in the initialization phase is reduced,and the resource consumption is also small.Then,for the random strong problem,this paper proposes a concept of label influence value.This method considers the selection of labels from three aspects,and it does not guarantee the stability of COPRA,so it is proposed on this basis The Label_Inf(l)method.Compared with the COPRA algorithm,the new algorithm has increased in time complexity,but it has greatly improved the stability of the algorithm,thus ensuring that the algorithm can mine high-quality overlapping community structures.In order to verify the feasibility of this algorithm,two benchmark datasets were selected,Zachary karate club and American College football and an artificially synthesized network dataset to test and evaluate the proposed algorithm.Comparing the algorithm of COPRA with this algorithm by overlapping the two community data sets,the algorithm proposed in this paper is improved in terms of the label initialization phase and the label selection phase.Although the time complexity has been improved,the modularity Q has also been improved.This shows that the algorithm proposed in this paper improves the stability in the community division and enhances the accuracy.Compared with SLPA,COPRA and this paper using artificial synthetic network datasets,we find that the decrease in NMI value of this algorithm is smaller than that of SLPA and COPRA.This indirectly shows that the algorithm can still be used in the case of complex community networks.Divide a better community.