Research On Retweeting Behavior Prediction Model Of Networks Based On Mixed Features

In recent years,with the popularity of mobile devices,the behaviors of netizens in social networks have became valuable information for analyzing user behaviors and human activities.Such as thumbs up,comments,retweeting and so on.This thesis mainly studies the retweeting behavior of users,which is considered as the main mechanism for information transmission in Sina weibo network.Retweeting prediction is of great significance in public opinion monitoring,topic detection,user influence assessment,advertising and microblog recommendation,etc.,that's why retweeting it has attracted extensive attention and research from a large number of scholars.In the actual decision-making process,whether the user is interested in the content of the tweets,the popularity of the tweets,and the influence value received by the user,all of which can affect the user's retweeting decision.The key to improve the accuracy of retweeting prediction is how to more comprehensively capture the multiple factors that influence users' behaviors and effectively combine them together.In this thesis,the mixed features are applied to prediction models,and the prediction process is discussed in stages by using the existing algorithms.The validity and applicability of the features and models are proved by the comparison experiments.The main content of this thesis including:1)The adoption of the under-sampling method based on microblog granularity to process unbalanced data.In this thesis,negative samples are sampled based on the granularity of microblog,which not only ensures the balance of positive and negative samples' size,but also ensures that the number of microblogs involved in negative samples is sufficient.In addition,the following experiment proves that the under-sampling strategy adopted in this thesis can make the model have better prediction effect than the training data set obtained by completely random under-sampling.2)The study of the factors affecting the retweeting behaviors of users,the analysis and extraction of mixed features.This thesis not only considers the relationship between users and users,users and microblogs,but also proposes the influence of circle on users' retweeting behavior from the perspective of "circle of friends".Besides,this thesis finds that users' own emotional preference also has a great influence on their retweeting behavior.3)A two-stage user's retweeting behavior prediction model is proposed.The prediction process is divided into two steps.The first stage predicts the retweeting probability of the user in single scene,and the second stage weighs the importance of each scene to the user's final retweeting decision.Compared with the prediction using classifiers directly,the two-stage model has better experimental results.4)The achievement of retweeting prediction model based on the mixed features.First,based on the features proposed in this thesis,we use different type of features and different algorithms to build our retweeting prediction models,then display and analyze the prediction results.And next we describe the importance of all kinds of features.Finally,this thesis carries out comparative experiments with the retweeting prediction models proposed by other papers.Compared with the mixed features proposed in the other two papers,the features proposed in this thesis show a better result on retweeting prediction problem in this thesis' data set.