Research On Semi-supervised Community Detection Method Based On Distance Dynamics And Anti-noise Performance

Community detection is an important problem that must be solved to understand,analyze,and control complex network systems.In recent years,scholars have put forward many methods for detecting communities in complex network.Most of these methods rely on network topology to perform unsupervised clustering of nodes and ignore the priori information that objectively existed,such as the community affiliations of nodes or community relationships between nodes which have be known.Community detection methods based entirely on network topology do not allow for higher accuracy,especially in the presence of noise information or blur community structure.Recently,many scholars have proposed some semi-supervised community detection methods,these methods attempt to fuse prior information to improve the performance and accuracy of community detection,but they tend to have high time complexity and are not suitable for dealing with large-scale complex networks.In response to the above issues,this article proposes a fast semi-supervised community detection method—SemiAttractor,by integrating the priori information and network topology into distance dynamics,the community detection process is accelerated under the influence of network topology and prior information,and the negative impact of noise on the community detection process is reduced,the community structure can be identified more accurately.The innovative results achieved are as follows:(1)A semi-supervised community detection method based on distance dynamics is proposed.Under the condition that the network community structure is not obvious.The basic idea of this method is: The entity in the complex network is equivalent to the particle.In the dynamics,each particle undergoes motion due to the force of stress and changes the distance between them.The process of balancing the various particles is equivalent to the process of community detection in complex networks.The distance between two nodes on each side of a complex network is influenced by their direct connections,common neighbors,and the individual neighbors of each node.It is equivalent to the influence of exclusion and attraction between the particles in the dynamics.Direct connections and common neighbors can narrow the distance between them.That is,there are more similarities between these two nodes.However,independent neighbors can increase the distance between them.Our approach can find communities of any size and canbe very useful in identifying small communities or abnormalities,the proportion of the same prior information is higher than other semi-supervised community detection methods.This method simulates the dynamic distance model between particles.that are typically found in real-world networks.This article has been tested in both artificial baseline networks and real-world networks to verify the effectiveness of our method.(2)A noise perturbation network model based on edge is proposed by deleting the real links and adding the wrong links.Without changing the average degree of the nodes,it meets the noise in the real network.In the noise perturbation network model mentioned above,we try to use the semi-supervised community detection method proposed in this paper and the current classical semi-supervised community detection algorithm to detect the community structure.From an experimental point of view,we analyze the impact of noise scale on semi-supervised community detection methods and verify the anti-noise performance of the semi-supervised community detection method proposed in this paper.