Biased sampling from facebook multilayer activity network using learning automata

Multilayer network simplification: Approaches, models and methods

“…The idea is to perform a two-stage random walk based sampling, where the first stage consists in selecting a relation type on which to walk (i.e., a layer), and the second one in enumerating the neighbors with regards to that relation only. More recently, Khadangi et al [60] addressed a similar sampling context taking Facebook as case in point, by proposing a biased sampling techinque for a multilayer activity network, where the activities are regarded as multiple social interactions (e.g., like, comment, post and share). The idea is to use a reinforcement learning scheme, i.e., learning automata [112], in order to learn transition probabilities among the users, and then apply a random walk-based sampling on the activity network using the learnt probabilities.…”

“…Two promising alternative approaches to the simplification process are based on node-layer relevance and model-based filtering; however, the design of such methods for multilayer networks is still in its infancy. Moreover, as concerns sampling approaches, the only methods for multilayer networks (i.e., [42] and [60]) belong to the exploration-based sampling subcategory, whereas no representative exists for the random access subcategory.…”