Modeling Method And Application Of Social Network User Influence

With the popularization of mobile devices and the emergence of social media,a huge amount of user data has been gathered to form a variety of social networks.How to extract key features from diversified information,and express user influence concisely and quickly,is conducive to situational awareness and public opinion orientation,and is helpful to regional comprehensive management.Although the research in this area has become saturated,the existing modeling methods still have the following problems: firstly,the problem of data default and untrustworthiness,including data missing,false information and privacy protection;secondly,the problem of incomplete information extraction,including single research perspective,complex and redundant features,and the similarity and heterogeneity between data;thirdly,the problem of influence differentiation,including different types of behavioral data,the potential distribution of user groups,and the assessment of differences in local areas,etc.This thesis mainly studies these three problems,and the main work results are as follows:Firstly,propose a general influence modeling framework to suit practical applications.This thesis divides the research status into four aspects including user and content characteristics,deep learning technology and node recognition.It analyzes the data,users,behaviors and influence factors in social networks,and also models influence based on user interaction information to form a unified expression framework.Next,aiming at the problems of data default and untrustworthiness and incomplete information extraction,propose a user influence evaluation method based on structural centrality(SDRank).This method considers information cascades and time nodes,combines the three-degree principle and the Page Rank algorithm,and further expresses the user's behavioral characteristics and event propagation characteristics,is suitable for social data sets with simple structures.Based on the Weibo data set and Twitter data set,this thesis verifies the effectiveness of the SDRank algorithm through comparative experiments with the Page Rank algorithm and the Trust Rank algorithm,and also analyzes some statistical characteristics of the relevant data set.Then,aiming at the problem of SDRank algorithm improvement and influence differentiation,propose a social network user influence model that integrates circles(LUIM).Based on the Louvain community algorithm,it combines the two concepts of social circles and user similarities to form a moderately large-scale collection of social relations circles;on the basis of the mapped user behavior network,through the random walk deformation algorithm and the original influence evaluation method SDRank,it further extracts the partial increment of the target user,thereby obtains the updated rank of influence users.Through comparative analysis,this thesis verifies that the model has good performance and optimizes the SDRank algorithm.Finally,introduces the specific application of the social network user influence model(LUIM).A social network user influence model integrated with circles,as a key module,realizes the user and event information query function,the user influence rank query function and the comparative analysis function.