Research And Application On Multi-diffusion Source Location Technology Based On Community Detection

Information source location is one of the most important research directions in the field of social network,and it is a key technical means to study source diffusion.For issues such as the location of the source of propagation,it is very important to study the source and the trend of propagation.However,due to the complexity of network nodes,it is difficult to explore the number of sources of propagation,as well as the direction and breadth of propagation.This makes the existing methods of tracing the source of propagation very difficult.It is difficult to apply to large-scale and complex networks;in addition,the previous methods are difficult to extract the depth representation of the latent network,and cannot automatically use enough parameters to balance different information sources.Aiming at the above-mentioned multi-source localization problem in social networks,this thesis combines the automatic extraction of deep features of deep learning,and proposes a method of spreading source point localization based on community detection.This method can not only provide a non-linear deep representation of the network,but also automatically learn the most appropriate balance of different information sources,providing a new solution for the traceability of social networks.The main contributions of this thesis include the following three parts:1.This thesis proposes the community detection method based on capsule network(CD-BOCN),its purpose is to mine the community structure in the network.This method first uses node features extracted from graph convolutional networks to generate highquality graph embeddings to integrate network topology and node content,and then use dynamic routing mechanisms to generate graph capsules and class capsules,and finally through class capsules for community detection.Experiments show that the method proposed in this thesis has good performance,which has great application value for the research of actual social networks.2.This thesis proposes the multi-diffusion source location method based on graph(BA-MPSL),its purpose is to find the original spreading source based on the existing community division.This method first generates a breadth-first tree from a social network with an obvious community structure based on the information recorded by the observation point,and then uses the maximum likelihood estimation criterion to calculate the probability of all candidate nodes as the source of propagation,so as to proceed to the diffusion source location.Experiments show that this method has a good performance in both the model network and the actual social network.3.In order to verify the effectiveness of the multi-propagation source location algorithm based on community detection proposed in this thesis,a multi-propagation source location system based on CD-BOCN and BG-MDSL was designed and implemented.The system can provide users with the functions of uploading citation network data sets,community detection and source location results display,and achieve the purpose of interacting with users through a visual interface.