Exploiting behaviors of communities of twitter users for link prediction

Link Prediction in Online Social Networks Using Group Information

“…In this work, we use the undirected and directed definitions as in [15] for five local measures: Common Neighbors (CN), Adamic Adar (AA), Jaccard Coefficient (Jac), Resource Allocation (RA) and Preferential Attachment (PA). Afterwards, they are referred to as base measures.…”

“…Thus, network information such as the structural ones and nodes attributes are used to build a set of feature vectors for linked and not linked pairs of nodes [5], [1], [15]. Classifiers are able to capture important interdependence relationships between nodes since feature vectors are formed based on unsupervised link prediction measures that capture different structural information sources of networks [9].…”

“…Furthermore, unsupervised evaluation measures are applied directly on results of link prediction measures but supervised evaluation measures are applied on results of classifiers [15].…”

“…Thus, Γ out (x) = {y | (x, y) ∈ E} is defined as outgoing neighborhood and Γ in (x) = {y | (y, x) ∈ E} is defined as incoming neighborhood [17]. Also, the set of all common neighbors of the pair of disconnected nodes (x, y) is defined as [15]. Considering these definitions and based on the approach showed in [13], we propose three new link prediction measures.…”

“…These methods can be divided into two different strategies: unsupervised [8], [13], [14] and supervised [5], [9], [1]. Furthermore, the strategy employed for a specific method influences how its performance will be evaluated [15].…”