Improved Label Propagation Community Discovery Algorithm And Visual Display

In today’s world,network science is an effective way to express the relationship between entity systems in many fields.In real life,the network has become ubiquity.For example,in the social field,network can be used to model the friendship and relationship between people;in biology,network are used to model the proteins needed for a function and the connections between them,so they can capture merabolis oprocesses in the organism;In the business field,we use the network to model the relationship between users and commodities,so that we can recommend commodities to users and improve their purchasing efficiency and frequency;in other domains,networks can also be used to model different relationships between different entities.In the network,the nodes represent different entities and edges represent relationships between entities.An important feature of complex networks is that entities tend to come together to form communities.Community detection can better demonstrate the structure,behavior,dynamics and organization of a complex network,thus it allows a better understanding of a network,and gives a deeper insight of interesting characteristics,that could not be caught if considering the network as a whole.The community detection algorithm has very important research significance and research value.In view of the problems existing in the existing community discovery algorithm and visualization platform,the main work of this paper is as follows:(1)Aiming at the label fluctuation caused by the label assignment process in the traditional label propagation algorithm and the randomness of the algorithm result caused by the randomness of the label update sequence,an improved label propagation algorithm based on random walk is proposed(A community detection algorithm based on random walk label propagation,LPASN).Firstly,the distribution of possible locations of random walks is used to measure the importance of nodes in the network,and the possibleprobability distributions obtained by each node are superimposed to obtain the importance of each node,then the importance of the nodes is performed in descending order.According to the importance of the nodes in descending order of the label update order of the nodes;Secondly,we can traverses the update sequence of nodes,compares each node with the node before it,calculates the similarity.If the node and the node before it are neighbor nodes and the similarity between nodes is greater than the threshold,then the node before it is sorted is selected as the seed node.Finally,by spreading the label of the seed node to the rest of the nodes,we can get the final partition result of the community generated by the LPASN algorithm.LPASN algorithm has better community division effect and is more stable than the original LPA algorithm.The experimental comparison with some classic algorithms on 13 real networks shows that LPASN has a better community division effect and performs well on indicators such as NMI,ARI,and modularity.(2)The visual display platform Network of complex network community discovery algorithm is designed and implemented.In the existing complex network analysis platform,users can only show the final experimental results of the algorithm,but cannot track the intermediate process of the algorithm and visualize the intermediate results,which will affect the accuracy of the algorithm improvement.In view of this phenomenon,we design and implement a visual display platform Network of complex network community discovery algorithm,which integrates different types of data such as social Network,biological Network,technical Network and so on,and provides a unified interface to the running algorithm.The system can display the intermediate clustering results of different algorithms so that users can improve the algorithm better.In addition,the system also allows users to upload the algorithm results,so that the system can analyze the algorithm results and calculate the indicators.Finally,two overlapping community detection algorithms based on geneticstrategy(i.e.A multi-objective particle swarm optimization algorithm and a link clustering based memetic algorithm)are compared and the intermediate steps are visualized on the platform.So that users can better understand the performance of genetic algorithm,and more targeted to improve the algorithm.