A positive example is a pair of nodes labeled as having the relationship that a link-prediction classifier should recognize. Existing relationships provide positive examples for training and testing. section 10.2.3 section 10.2
Positive examples support link-prediction workflows in several ways:
- Pair with negative examples so the classifier can distinguish relationships that exist from relationships that do not. section 10.2.3
- Supply relationship cases for training and evaluation while network features are calculated from a separate feature set. section 10.2
- Help control class imbalance because possible non-relationships can greatly outnumber existing relationships in sparse graphs. section 10.2.3
Counting positive examples
A link-prediction exercise produces 13,082 positive examples. A practical workflow selects approximately the same number of negative examples, then keeps the relationship data organized across feature, training, and test sets. section 10.2.3 section 10.6
