Research On Signed Network Community Detection And Signed Prediction Based On Balance

Birds of a feather flock together,people flock together,societies and categories are concepts that are relevant to people's lives.This situation is no exception in the rapid development of information technology today:a wide variety of goods,new ideas of a hundred to promote the emergence of different communities.There is a strong correlation and similarity between the individuals in the community.Based on the characteristics of nodes,the network communities can be accurately divided,and the set of nodes with similar characteristic labels such as interests and hobbies can be obtained,which is of great significance for business decisions such as commodity classification and recommendation.At the same time,the community structure can also help us to predict the unknown information in the network,which is also helpful to quantify the potential risk of the financial network and predict the evolution of the network.The development of social media has produced a large amount of emotional interaction information with positive and negative attitudes(such as evaluation,attention,thumb up,shielding,etc.),which has helped to comprehensively and deeply explore the connections between individuals.Signed network is a powerful tool to represent this kind of information.In the real signed network,the generation of groups is not only affected by the affinity between individuals,but also by the attitude between individuals,the degree of network evolution and other factors.In the work of community partitioning,we explore the mechanism of the network equilibrium module,and quantify the contribution of the nodes in the network to the local structural balance based on the structural balance theory of signed networks.This paper holds that the more positive influence a group of nodes has on the formation of equilibrium structure,the more stable the group will be,and the more supportive the relationship between these nodes will be,and the stronger the intention of information transmission and opinion transmission among them will be.On this basis,we combine the balance contribution value with community detection,and extend the classical modularity algorithm to the signed network to mine the balanced community structure in the signed network.Then,based on the above signed network balanced community detection algorithm,the effectiveness of the algorithm is verified by numerical simulation on the artificial network.In the real emotional interaction information network,negative information(such as rejection,hostility,suspicion,etc.)has a negative impact on the formation of the collective,and the collective will gradually lose unity and collapse with the increase of negative information.On the contrary,positive information plays a positive role in this aspect.Therefore,we discuss the effect of the algorithm by changing the number of positive and negative edges of the network and the density of the connecting edges of the nodes in the community.Balance evaluation index(BS)and similarity index(S)are introduced to quantify the degree of balance of the network and detect the similarity between the community and the original community.Compared with the classical DM algorithm(proposed by Doreian and Mrvar),the algorithm has a more accurate detection effect when the structure is clearly collectivized.The reasons for this result are as follows:the more interactive information between individuals in the network,the more intimate the relationship between them,and the easier it is to be detected into the same community.In addition,when the signed information is chaotic,the detection effect of the algorithm is better than that of the DM algorithm.This is because the algorithm fuses the signed information with the structural information and abstracts the network local information to contribute to the small balance module.This makes the algorithm insensitive to the change of signed information and has better robustness.Finally,the algorithm is also applied to two signed networks with real community structures,and the results show that the communities detected by the algorithm can match the real communities completely.To sum up,our detection algorithm provides an effective solution for mining signed network communities.We also focus on the prediction of unknown and missing signs in the signed network.Whether it is the potential friend mining in the social network or the preference recommendation in the commodity selling network,it is necessary to predict the sign.Completing the network of positive and negative information is also of great significance to the dynamic process such as the positive propagation of public opinion.Therefore,a sign prediction algorithm is proposed in this paper.In view of the artificial simulation model of the signed network with community structure,the variation of accuracy predicted based on the balanced community detection results in the network with different structural parameters is discussed.The changes of network structure and the amount of original positive and negative information have a certain influence on the prediction effect.In addition,the changes of the algorithm under different training sets are discussed.The results show that the algorithm has the best performance when the positive edge information is more and the degree of collectivization is stronger,and the network balance is higher.In this case,the algorithm will not become significantly worse because of the smaller size of the training set,and has good robustness.All these conclusions can provide some ideas for product recommendation,friend association and other work.