Research On Retweet Predicting Based On Social Network

With the rapid development of information technology, the scope ofapplication is more and more widely of broadband network and Mobile InternetDevice, promote the speed and frequency of the exchange between people greatly.The advent of social media，brought change to the people exchange with eachother. With ultra-high user participation, the social media make a large number ofusers together in the social networks with the rapid rise of microbloggingplatform, the amount of information also rapid expanding in the social networks.Master the propagation of information can have a significant impact on the earlywarning of public opinion and the enterprise publicity.Therefore, predicting the spread of information in social network scalebecomes a hot issue of current research.Retweet predicting adopts predicting the forwarding times of information tospeculate about the target information dissemination size in social networks. Inthe current research in the field of retweet predicting, there are two differentresearch angles, based on the Information propagation tree modeling and basedon machine learning theory modeling. Because the latter can handle hugeamounts of data, then we choose this Angle were studied. The research content ofthis paper is divided into the following sections:First of all, according to the results of the survey，we choose the user feature,text feature, temporal feature and metadata feature as the model feature. We focuson the extraction method of the text topic feature based on LDA topic model. Inour experiments，prediction model has better prediction effect when the LDAtopic model’s topic number equal to80. Secondly，in view of the time complexity and space complexity is very highto establishment of an Information propagation tree, proposed that make theinformation of social networks serialization. We have established the binomialclassification prediction models based on Logistic regression and Support VectorMachine, and conducted a parallel processing because the training set dataquantity is too huge. Experiments show that the two kinds of prediction modelsboth has good prediction effect, and the prediction effect of the prediction modelbased on Logistic regression is better than the model based on Support VectorMachine.Finally, in order to predict the propagation scale of information in socialnetworks more accurate, this paper presents a method to establish the regressionprediction model based on regression analysis method. Then we can predict theforwarding number of target microblog information directly. We have establishedthe multi classification prediction model based on Logistic regression and theregression prediction model based on Ridge regression. In the experiments ofpredicting the forwarding number of information, the Mean Squared Error ofregression prediction model reduced29.44%than the multi classificationprediction model，the method that for retweet predicting based on regressionanalysis is effectiveness.