Semi-supervised Learning Community Detection Algorithm Based On Particle Competition

Community detection is of extraordinary significance in comprehending the structure and functions of complex networks.A plethora of exhaustive studies have proved that community detection methods based only on topology information tend to obtain poor community partition results.Several methods that utilize semi-supervised learning and prior information to improve performance are proposed.The particle competition algorithm is a quick and heuristic semisupervised learning algorithm when applied to community detection.However,the existing particle competition algorithms have a series of challenges and needs to be further improved.Details as follows:1.The algorithm does not make full use of all the information of the network,and has disadvantages such as poor robustness,poor stability,and low accuracy.Furthermore,it cannot be effectively applied to overlapping community detection.2.The algorithm ignores the influence of the noise of prior information.Prior information can be uncertain,imprecise,or even noisy.Algorithms rely too much on prior information,which may cause wrong prior information to spread throughout the network,thereby reducing the effectiveness of community detection.3.The algorithm cannot effectively use pairwise constraints,and the algorithm needs to know the number of communities in advance.In addition,it is difficult for algorithms to realize community detection tasks on networks with fuzzy or unbalanced community structures.In view of the above situation,this paper focuses on the following three aspects to carry out research on community detection in complex networks based on semi-supervised particle competition:1.A semi-supervised overlapping community detection algorithm based on particle competition is proposed.The algorithm divides the community formation process into initialization phase,walking phase,restart phase,convergence phase and overlapping community detection phase.The algorithm not only has high accuracy and low time complexity,but also can be effectively applied to overlapping community detection tasks.2.A community detection algorithm in error-prone environments that combines dynamic distance and particle cooperation and competition is proposed.The algorithm sets up a mislabeled vertices detection rule,which not only improves the accuracy of community detection,but also has good robustness in error-prone environments,which can detect and prevent the spread of mislabeled nodes.3.A four-stage community detection algorithm that combines particle cooperation and competition and graph entropy is proposed.This algorithm improves the particle cooperation and competition algorithm so that it can use pairwise constraints and reduce randomness to obtain good community partition results.Furthermore,by combining graph entropy and optimization strategies,the algorithm can be applied to networks with fuzzy or unbalanced community structures.Crucially,the algorithm automatically determines the number of communities.