Research And Application Of Information Dissemination And Link Prediction In Social Networks

The structure and evolution of social networks have important implications for social,economic and political aspects.Identifying the sources of information dissemination and important centres of information aggregation in social networks,and predicting links and behaviours in social networks are conducive to better understanding the interaction patterns between users and information dissemination patterns,as well as solving problems related to information dissemination and public opinion control in social networks.This thesis focuses on two key technologies of social network key node identification and link prediction,and the main innovations and works are as follows.(1)A key node identification model based on graph neural network is proposed.Many common key node identification algorithms are based on a node-in-the-centre approach.To address their problems of single features and poor generalisation,this model extracts the potential structure information of nodes through a graph embedding algorithm,and then uses the obtained node feature matrix and adjacency matrix together as the input to the graph attention network for learning the representation of nodes.The influence labels of the nodes are obtained through SIR simulation experiments to transform the key node identification task into a supervised learning task.Through comparative experiments,it is demonstrated that the accuracy of the key node identification model proposed in this thesis is significantly better than that of the centrality-based approach,and can accurately identify key nodes in social networks.(2)A link prediction algorithm incorporating subgraph segmentation and gated attention mechanism is proposed.To address the problems of high computational complexity of graph neural networks for large-scale social network datasets and the difficulty of message aggregation and updating,this algorithm simplifies the computation of large networks to that on neighbourhood subgraphs centred on predicted links through subgraph segmentation and subgraph structure information extraction.By aggregating subgraph information through a gated attention mechanism,the relationships between nodes can be captured more accurately.Experimental results demonstrate that the link prediction algorithm constructed in this thesis can perform the task of social network link prediction well and the model is highly accurate.(3)Based on the key node identification and link prediction algorithms proposed in this thesis,a social network information dissemination and link prediction analysis system was designed and implemented,which consists of six modules,including data management,system management,key node identification and link prediction,etc.After functional and performance tests,the system functions met expectations and performed well.The key node identification method and link prediction method implemented in this thesis both effectively improve the accuracy rate and other indicators,enhance the recognition and prediction effect of related tasks in social networks,and have certain reference value for social network related research.The implemented social network analysis system has certain application value in the fields of public opinion management and social relationship analysis.