I am trying to solve the link prediction problem in a directed graph using supervised learning. In the case of an undirected graph, it is pretty straightforward. For instance, first, compute the embeddings and solve the problem as a via binary classification problem, where, node embeddings need to be transformed into edge embeddings by averaging or adding ‘i’ and ‘j’ node embeddings. This makes sense in an undirected graph because i ->j and j->i edges represent the same link. However, in a directed graph, it doesn’t make sense to transform the node embedding pair into edge embedding by adding or average. Can anyone help me out on how I can convert the node embedding into edge embedding?
Looking forward !