The Influence Prediction Algorithms In Social Networks

With the development of the Internet,more and more people choose to distribute information and share resources through online social network sites.The propagation of information can be maximized through the influences among individuals.Researches on the influence among individuals are of great use in marketing,sociology and politics etc.The thesis mainly studies the standards for evaluating the influential members and the ways for predicting the transmission of information,then it realizes the prediction of member promotion and posts popularity respectively.The main work of our thesis is as follows.(1)For the problem of potential influential member promotion in social networks,we propose the concept of "ReputationRank" by referring to the "PageRank" algorithm of calculating the importance of web pages by Google and add it as the third attribute for measuring the influence of a member.What's more,the other two attributes are Activeness(out-degree)and Popularity(in-degree).Then the skyline operator is introduced for solving this multi-decision making problem.We consider the Infra-Skyline as our candidates when predicting the potential influential members.Then in the process of member promotion,we put forward the concept of skyline distance.By calculating the value of skyline distance,we obtain the necessary condition of not being dominated.At the same time,the searching space of promotion plans can be remarkably reduced by cost-based and domination-based pruning strategies.Finally,the comparison experiments on the real datasets such as DBLP and WikiVote prove the effectiveness of the algorithm.(2)For the problem of information propagation in social networks,we propose the algorithm PMKF for predicting the popularity of the post by referring to the idea of the Kalman filter,and we consider the total number of forwardings as the measure of post popularity.We firstly predict the influence of the post.This process includes two steps.The first step is prediction.It makes use of the optimal estimate of the post influence in the last moment to make a prediction of the influence in the current moment.The second step is correction.By observing the forwarding process,we can modify the predicted value in the last step and obtain the optimal estimate.Finally,by building a future information cascade tree,the prediction of the total number of the forwardings can be transformed into the process of a geometric sequence summation.Thus,the prediction of posts popularity can be realized.Experimental results on real datasets show that PMKF can predict the popularity of posts effectively.