Research On The Impact Diffusion Of Social Network Events Based On Group Multimodal Characteristic

With the development of Internet technology,social network has become more and more closely connected with people’s life.Various events are attached to social networks and their influence spreads rapidly on social networks.Propagation relies on various user nodes in the network.When an event occurs,user nodes in the social network will have their own views.Users have a strong recognition of their own views,and at the same time,they will continue to spread their views in their own way in the social network.Add fuel to the flames on the impact of communication events.Existing research usually analyzes the social network as a graph,which mainly analyzes the spread of event influence based on the topology and interaction characteristics of nodes in the network.However,due to the exponential increase in the number of user nodes today,the research relies on a single node feature mode.Existing research often ignores the key factor of the group community phenomenon that develops over time in social networks,which leads to the inability to accurately analyze the spread of influence.In this regard,this paper uses the multimodal characteristics of nodes in social networks and combines the phenomenon of group communities to propose a social network event impact diffusion analysis framework based on group multimodal characteristics.The specific research work is as follows:First,a multi-modal feature fusion model for user nodes is designed.The web crawler is used to crawl Weibo users,and the multi-modal feature information about events in the user’s latest blog posts and the interaction information between users are extracted.Consider users as nodes and the interaction between users as edges to construct a social network graph.For the social network graph,the number of interactions between users is used as the weight of the link edge,and for the node text features,a network new word discovery framework based on sentence semantic similarity is designed to analyze the text and fuse it with the node image features as the node’s multi-mode state attributes.Secondly,we design the core level division algorithm of social network groups and communities.For the constructed social network graph,framework rely on the Louvain algorithm to divide the groups and communities,and clarify the division of each group and community.At the same time,a graph degeneracy framework based on weighted attribute network is designed.In this framework,the node social environment is introduced,and the core level is divided based on the attribute weight in the social environment.This framework is used to divide the core hierarchy of each group community,analyze the hierarchical structure of each group community subgraph,and extract the corresponding core subgraph.Finally,an event influence diffusion model based on graph embedding expression is designed.A graph kernel graph neural network model with dual neighborhood aggregation is designed,and the divided core subgraph is used as input,combined with node attributes and topological structures in the core subgraph,to construct a corresponding Graph embeddings expressed as population multimodal features.By analyzing the similarity relationship of group multimodal features,a group influence resonance algorithm based on graph similarity is proposed to clarify the core categories(activation and non-activation)of each group.Afterwards,an influence diffusion algorithm based on the core sub-graph level is designed to achieve the purpose of analyzing the influence diffusion of social network events.In order to verify the effectiveness of the framework,we compare the node multimodal fusion part and the graph degeneracy framework part with the existing methods,and compare the dual neighborhood aggregation graph kernel graph neural network model with the existing graph neural network and basic graph Kernel function and other methods for comparison.The experimental results show that this research framework has achieved good analysis results,which verifies the feasibility of this research.The framework of this paper can provide a theoretical basis for realworld work such as regulating public opinion,tracing rumors,etc.,and has good practical application significance.