A Research On Rumor-publishers Recognition Of Microblog Oriented To Network Public Opinion Control

With the coming of Web 2.0,social media such as Twitter and microblog have developed rapidly.Social media service not only gives people a way to get information conveniently and quickly,but also enrich our entertainment life.Because of its liberalization of expression,diversification of information and rapid dissemination,it has become a hotbed for rumors to spread.The quick spread of rumors disturbs our daily life and threatens social security.Therefore,the identification of social network rumors has attracted the attention of scholars.The researchers wish to prevent,monitor and manage the network rumor effectively through the research on automatic detection of social network rumors.Many scholars at home and abroad have carried out mature research results in the research field of social network rumor identification.However,the root of the spread of online rumors lies in the existence of rumor publishers.While there are few study on the rumor publishing accounts,most of them just regard the attributes of rumor publishers as a feature for rumor identification task.We think the identification of the rumor-publishers is the same importance,so we proposes to take the rumor publishers identification as the research object.We hope to cut off the produce and spread of social network rumors from the root,which is of great significance in theory and practice.The social network rumors affects people's life and interferes people's judgment of things.However,for the purpose of seeking self-interest,self-expression and soliciting attention,these rumor publishers still publish many fake news on social network platforms which cause people's negative emotions and threaten social stability.To solve this problem,we collect 3980 rumor microblog and the corresponding publishers' information and 3514 normal account information.We use these data to build the dataset for this experiment.Then,the behavior characteristics of rumor publishers are analyzed.From the statistic results,we extracted ten features to construct the feature of rumor publisher identification.Next,we use XGBoost algorithm,Support Vector Machine algorithm and Naive Bayesian algorithm to build classifier.The experimental results show that both XGBoost and SVM are effective in rumor publisher recognition.The accuracy of classification can reach more than 80%,but the effect of Naive Bayesian is poor.Finally,considering the actual situation,we test the performance of classifiers on unbalanced datasets.The results show that the classification performance of the three classifiers decreases with the increase of unbalanced degree of datasets.The XGBoost classifier always has the best performance among the three algorithm.