Not all the preceding metrics can be used at the same time. Which one is best to use in which situation depends on the underlying process governing the graph growth and, in many situations, it is necessary to test several metrics to find the most appropriate one. We can even imagine more metrics, such as the following:
- Reciprocity: The presence of a link makes the addition of a link in the opposite direction more likely and the removal of a reciprocal link less likely.
- Newness weakness: Newly formed links are less likely to persist than older links and hence hold less weight.
- Instability: If the properties or links attached to nodes u and v change very often, the edge between u and v is also more likely not to survive and has a less weight.
Following this review about link prediction scoring methods, we are now going to build a link prediction model using Neo4j and scikit-learn.
