Competition And Cooperation Model For Information Diffusion In Social Network

As brand-new social concepts of opening,sharing and interaction are accepted by users all around the world in the recent years,the online social network platform has replaced the traditional media and integrates into every aspect of our lives.As for the social network platform,it is important to keep large amount of active users and hits in order to maintain the development and profit.So the key point is to attract users and to satisfy their needs.And the hot topic prediction and recommendation is one of the core technologies.This paper focuses on the prediction of information diffusion in the social network.According to the analysis of hotspots in Weibo,we discover that messages with strong relationship will affect each other and show phenomena of interaction and competition.However,many previous works have overseen this influence and take the diffusion process as a self-growth process.According to the summary of related works,the diffusion predictive models could be divides into two categories: activation model based on the node and the predictive model based on the network.The activation model focuses on analyzing the process of how a specific node is affected by its neighbors.ICM and T-Basic models belong to the activation model.But this kind of models neglects the interaction and competition of messages,and has not considered that the force of influence could fade.So in this paper,we put forward an improved activation model.Besides,the predictive model based on the network focuses on the diffusion rate and amount,taking SIS-based model and PDE-based model as examples.However most of these models also oversee the interaction.In this paper we combine these two models,and get the Coco model from the derivation of the improved activation model.The Coco model has low computation complexity and is based on the reasonable characteristics of the social network.We crawl the dataset from Aug.2014 to Dec.2015 in Xinlang Weibo,and the dataset contains 1,263,988 microblogs and 905,987 users' information,including hotspots like the Ice Bucket Challenge and the wedding of Xiaoming Huang and Ying Yang.11 events of this dataset are selected to be used to evaluation the Coco model and others.The evaluation contains the fitting precision evaluation and the prediction precision evaluation.The fitting evaluation focuses on analyzing whether the model could explain the characteristics of information diffusion and match the real process well,and Coco model shows a great performance over other models.Besides the prediction evaluation is a key index to evaluate the significance of predictive model in a real social network.And the Coco model also shows great predictive precision and robustness.It does not have the problem of overfitting and the prediction offset is under control in an error range.In brief,the Coco model has advantages of low computation complexity,high fitting and prediction precision,and it could explain the phenomena of interaction and competition well.Therefore,it could be used in a real social network platform in the module of hot topic prediction and load balancing.