Research On Key Technologies Of Social Network User Characteristic Analysis

The social network is a platform that allows people to show themselves,establish contacts,communicate and interact,and express opinions.Its diverse information,mas-sive users,and convenient news make social networks an indispensable part of most peo-ple’s lives.As the main body of the social network,users are both consumers and produc-ers of information.It continuously generates user characteristic information from account registration to use on social platforms,including user attribute information,social link re-lationships,social speech content,and social interaction behaviors.This user characteris-tic information is important information for identifying the user’s identity,describing the user’s characteristics,and portraying the user’s portrait.Many web application services use this information to provide services such as personalized recommendations,emer-gency response,and real-time medical analysis.Therefore,the analysis of social network user characteristics has important practical significance and application value.However,the analysis of social network user characteristics faces many difficulties and challenges.On the one hand,the lack of user attribute information in social networks,false and difficult to distinguish,and inconsistent formats,especially the lack of user lo-cation information,make it impossible to carry out many network personalized services,such as local news push,disaster emergency response,etc.On the other hand,due to user privacy settings in social networks,the structure of social networks is incomplete,coupled with the large scale of users,diverse information,and complex relationships in social networks,how to effectively model on large-scale networks and use all aspects of user information to Predicting unknown links is a challenging task.Faced with the above-mentioned difficulties and challenges,this article starts from two aspects of user attribute information and user relationship information,and carries out two types of research respectively on user location identification and user link relationship discovery.In terms of user attribute information in social networks,in response to the scarcity of users’ open geographic information in social networks,this paper proposes a Hetero-geneous graph Attention network for user Geolocation(HAG).This model introduces an attention mechanism to combine valuable clues in social interaction behavior and social text,and integrates the characteristics of the two for location recognition.When dealing with social interaction behaviors,this paper creatively applies heterogeneous graphs to model various social interaction behaviors and introduces the heterogeneous graph atten-tion network(HAN)to learn network structure information.When dealing with social text,this paper proposes a contextual attention network(CAN),which uses bidirectional GRU to learn the contextual information of the text,and combines the attention mech-anism to focus on extracting geolocation relevant text information.Experimental tests are carried out on three public Twitter data sets.Compared with the existing benchmark methods,the method proposed in this paper shows the most excellent performance.In terms of user relationship information in social networks,in order to solve the problems of incomplete social network structure and large number of users,this paper proposes an Associated Subgraph Feature Fusion for user link prediction(ASFF)model.The model is trained in batches by extracting node association subgraphs.In the extracted association subgraph,this paper uses the double-radius node labeling method to mark the network structure of the nodes,which is used to enhance the network structure represen-tation ability of the model.In the feature fusion stage,this paper introduces a multi-head graph attention network to aggregate the node attributes and network structure attributes of multi-order neighbors,uses the attention pooling layer to aggregate all node features,and generates associated subgraph embeddings to predict whether there are unknown links between users.Experiments are carried out on seven commonly used link prediction data sets.Compared with the existing benchmark methods,the method proposed in this paper shows the most excellent performance.