Evaluating Connection Quality between Two Individuals in Social Networks

Along the development of social networks, predicting the quality of connections between two individuals has been attracting research interest from both industrial and research community. For many social applications with the prediction requirement, the personalities (labels) of individuals within the relationship (connections) were known to have a significant impact against the analysis, observing two neighbored individuals acquiring an identical label are more likely to affect each other than those are only with different labels. In the paper, we tackle with a more practical problem in which an individual has several labels, rather than considering only an identical label for all individuals. For the task, we first present multi-label independent cascade (MIC) model which is in fact a generalization of the classical IC model, and then in a multi-label social network, we propose two algorithms for analyzing the maximum connectivity between two individuals. The first is a heuristic algorithm, which, generalizing the maximal flow algorithm, is based on repeatedly finding shortest augmenting path; the second is based on linear-programming (LP) and produces optimum solutions. At last, we evaluate the two algorithms by experiments through their practical performance and runtime.