@MASTERSTHESIS\{IMM2016-06948,
    author       = "J. S. Huisman",
    title        = "Theoretical and Computational Analysis of Dynamics on Complex Networks",
    year         = "2016",
    school       = "Technical University of Denmark, Department of Applied Mathematics and Computer Science",
    address      = "Richard Petersens Plads, Building 324, {DK-}2800 Kgs. Lyngby, Denmark, compute@compute.dtu.dk",
    type         = "",
    note         = "{DTU} supervisor: Sune Lehmann, sljo@dtu.dk, {DTU} Compute",
    url          = "http://www.compute.dtu.dk/English.aspx",
    abstract     = "Big data has allowed to study the communication and mobility patterns of humans with ever greater resolution. However, it is not yet clear how information from online social networks relates to offline face-to-face interactions. Knowledge of this directionality will help us to harness the increasing wealth of online information to improve predictions of for example offline disease spread.

This study aimed to distill relevant information from online social networks to predict meaningful face-to-face contact behaviours. To this end data from the Copenhagen Network Study on the Facebook and face-to-face interactions of 850 students was used. First, a network of offline interactions was predicted using binary link prediction on features distilled from the Facebook interaction data. The network predictions of this model were then validated, using simulations of disease spread and comparison against the Erd¨os-Renyi random graph and configuration model network. It was found that stringent variables of offline contact, such as meeting during off-hours or meeting more than 5 times per week, could be predicted with 69\% accuracy, which was 19\% better than the Majority Vote Classifier. The target variable of meeting at least once a week could be predicted with 78\% accuracy. The predicted network showed disease simulations that closely resembled those on the actual network, and performed significantly better than simulations on the Erd¨os-Renyi random graph. 

To the knowledge of the author, this is the first study to validate the quality of the network structure resulting from link prediction using disease simulations. It was shown that online network information can be used to predict offline contact networks which are useful for the investigation of the spread of disease. This study paves the way for future verification of disease models and development of intervention strategies using primarily online network information."
}