authors : Tiago Pimentel1, Adriano Veloso1 and Nivio Ziviani1,21) using each node as root once
2) skipgram learn a model by maximizing the log probability
(1)
the probability of the model in this paper
Vt : a vertex in the graph
Vt+1 : another vertex in the graph : the representation of Vi
P(Vt+1|Vt , r) can be written as
(2)
: the transposed output feature vector of vertex j used to make prediction
are learned simultaneously by optimizing (1)