Are We Really Friends?: Link Assessment in Social Networks Using Multiple Associated Interaction Networks

Mohammed Abufouda, Katharina Anna Zweig

Abstract: Many complex network systems suffer from noise that disguises the structure of the network and hinders an accurate analysis of these systems. Link assessment is the process of identifying and eliminating the noise from network systems in order to better understand these systems. In this paper, we address the link assessment problem in social networks that may suffer from noisy relationships. We employed a machine learning classifier for assessing the links in the social network of interest using the data from the associated interaction networks around it. The method was tested with two different data sets: each contains the social network of interest, with ground truth, along with the associated interaction networks. The results showed that it is possible to effectively assess the links of a social network using only the structure of a single network of the associated interaction networks and also using the structure of the whole set of the associated interaction networks. The experiment also revealed that the assessment performance using only the structure of the social network of interest is relatively less accurate than using the associated interaction networks. This indicates that link formation in the social network of interest is not only driven by the internal structure of the social network, but also influenced by the external factors provided in the associated interaction networks.