Embedding-based methods
In this section, we describe a more advanced way to perform link prediction. The idea behind this approach is to solve the link prediction problem as a supervised classification task. More precisely, for a given graph, each couple of nodes is represented with a feature vector (
), and a class label (
) is assigned to each of those node couples. Formally, let 
be a graph, and for each couple of nodes 
, we build the following formula:
Here, 
is the feature vector representing the couple of nodes 
, and 
is their label. The value for 
is defined as follows: 
if, in the graph G, the edge connecting node exists; otherwise, 
. Using the feature vector and the labels, we can then train an ML algorithm in order to predict whether a given couple of nodes constitutes a plausible edge for the given graph.
If it is easy to build the label vector for each couple of nodes, it is not so straightforward to build the feature space. In order to generate the feature...
