Retweeting Behavior Prediction In Social Networks

Social network is an online community created by Internet,which provides ways for people to communicate.No matter in China or in foreign countries,it has become important part of people's life.Research on social network also attract researchers‘interest because of its massive scale real-world data and practical significance.Retweeting behavior is the core function of social network.It promotes information diffusion in social network.Research on retweeting behavior is not only beneficial for understanding the mechanism of information diffusion,but also helpful for pushing personalized message to users.Meanwhile,because of variety of users and message,it is challenging to predict retweeting behavior correctly.Massive scale of data also result in big difficulty.Retweeting behavior prediction is a difficulty and challenging task.This paper proposed Online Matrix Factorization Model(OMFM)to predict retweeting behavior.Firstly,it represent retweeting behavior in social network by an retweeting matrix.Secondly,it compose retweeting matrix into a latent user matrix and a latent message matrix.Finally,multiply this two metrics to reconstruct retweeting matrix,which is the predicted result.To solve two main problem of current MF model,OMFM takes 3 measures: First,n the loss function,time-decaying weight are assigned to samples to emphasize the contribution of new messages.Second,In the regularization part,only necessary elements in the matrix will be updated.Third,In the training part,initialize a new round with the solution of the old message.We evaluate our model on real-world dataset Twitter.It is compared with other current models in both prediciton performance and running time.The result shows the effectiveness of OMFM and it is obviously best choice for retweeting behavior prediction task.