MODELING INFORMATION DIFFUSION OVER SOCIAL NETWORKS FORTEMPORAL DYNAMIC PREDICTION

 

ABSTRACT:

How to model the process of information diffusion in socialnetworks is a critical research task. Although numerousattempts have been made for this study, few of them cansimulate and predict the temporal dynamics of the diffusionprocess. To address this problem, we propose a novel in-formation diffusion model (GT model), which considers theusers in network as intelligent agents. The agent jointly considersall his interacting neighbors and calculates the payoffsfor his different choices to make strategic decision. We introducethe time factor into the user payoff, enabling theGT model to not only predict the behavior of a user butalso to predict when he will perform the behavior. Both theglobal influence and social influence are explored in the timedependentpayoff calculation, where a new social influencerepresentation method is designed to fully capture the temporaldynamic properties of social influence between users.Experimental results on Sina Weibo and Flickr validate theeffectiveness of our methods

EXISTING SYSTEM:

dAta-centric models are usually learned from actual in-formation diffusion data, and can be divided into macromodelsand micro-models. Macro-models can generatediffusion cascades whose macro properties are similarto that of actual diffusion cascades. But they still can notpredict the information diffusion process. This limitationis addressed by micro-models, which can predict whether auser in a social network will be activated by a piece of information.Since the information diffusion process is causedby user behavior, information diffusion prediction is actuallyuser behavior prediction. Most micro-models can predict the behavior of a user, but they can not predictwhen the user will perform the behavior.PROPOSED SYSTEM:we propose a novel information diffusionmodel (GT model) for temporal dynamic prediction. Incontrast to traditional theory-centric models, the GT modelregards the users in the network as intelligent agents. Itcan capture both the behavior of individual agent and thestrategic interactions a these agents. By introducingthe time-dependent payoffs, the GT model is able to predictthe temporal dynamics of the information diffusion process.Different from most data-centric models, the GT model cannot only predict whether a user will perform a behavior butalso can predict when he will perform itWe make the followingcontributions in this work:• We propose a novel information diffusion model (GTmodel), where,between different choices (behaviors), theuser jointly considers all his interacting neighbors’ choicesto make strategic decisions that maximizes his payoff.• We propose a time-dependent user payoff calculationmethod in the GT model by exploring both the globalinfluence and social influence.• We propose a new social influence representation method,which can accurately capture the temporal dynamicproperties of social influence between users.• We conduct experiments on Sina Weibo and Flickrdatasets. The comparison results with closely relatedwork indicate the superiority of the proposed GT model.

CONCLUSION

We have presented a novel information diffusion modelin this paper. It regards the users in a social network asintelligent agents, and jointly considers all the interactingusers to make strategic prediction. By introducing the timedependentpayoffs, the model has the capability to predictthe temporal dynamics of information diffusion process. Boththe global influence and social influence are explored for userpayoff calculation, where the social influence representationmethod is newly designed for fully capturing its temporaldynamics. Experimental results have confirmed the rationalityand effectiveness of the proposed model.

REFERENCES

[1] E. Bakshy, J. Hofman, W. Mason, and et al. Everyone’s aninfluencer: Quantifying influence on twitter. In WSDM,pages65–74, 2011.

[2] S. Brin and L. Page. The anatomy of a large-scale hypertextualweb search engine. Computer Networks and ISDN Systems,30:107–117, 1998.

[3] M. Cha, A. Mislove, and P. K. Gummadi. Ameasurement-driven analysis of information propagation in theflickrsocialnetwork.InWWW,2009.

[4] W. Chen, C. Wang, and Y. Wang. Scalable influencemaximization for prevalent viral marketing in large-scale socialnetworks. In KDD, pages 1029–1038, 2010.

[5] W. Chen, Y. Wang, and S. Yang. Efficient influence maximizationin social networks. In KDD, pages 199–208, 2009.

[6] P. Domingos and M. Richardson. Mining the network value ofcustomers. In KDD, pages 57–66, 2001.

[7] H. Fei, R. Jiang, Y. Yang, and et al. Content based socialbehavior prediction: a multi-task learning approach. In CIKM,pages 995–1000, 2011.

[8] A. Goyal, F. Bonchi, and L. Lakshmanan. Learning influenceprobabilities in social networks. In WSDM, 2010.

[9] J. Leskovec, M. McGlohon, C. Faloutsos, and et al. Cascadingbehavior in large blog graphs. In SDM, pages 202–209, 2007.

[10] L. Liu, J. Tang, J. Han, and et al. Mining topic-level influencein heterogeneous networks. In CIKM, pages 199–208, 2010.

[11] D. Wang, Z. Wen, H. Tong, and et al. Information spreading incontext. In WWW, pages 735–744, 2011.