A Gibbs Sampling based method for collective classification in multilayer social networks
Abstract
This last decade has witnessed a rise in the interest for methods that combine both features and interactions in a dataset. Such methods are referred to as collective classification methods. Although their popularity is increasing, they are largely underrepresented in comparison with methods that use only features or only interactions, and despite the availability of datasets that contain both modes. This study proposes a collective classification method that aggregates both structural and attribute features from an element and its neighborhood. It defines the neighborhood of elements in a multilayer network such that neighbors are more likely to belong to the same class. It then learns a model using features of an elements and of its neighbors Finally, a variation of the Gibbs sampling method for collective classification is performed using varying neighborhoods.