Link Prediction: A Primer

Synonyms

Edge prediction; Relationship extraction; Social ties inferring

Glossary

Common Neighbors:

For two nodes, u and v, the set of their common neighbors is defined as N = Γ(u) ∩ Γ(v), and correspondingly the size of set N is |Γ(u) ∩ Γ(v)| (Newman 2001)

Jaccard Coefficient:

Jaccard coefficient (Liben-Norwell and Kleinberg 2007) is a normalization of the common neighbors metric, which is defined as $JC(u, v) = \frac{|\Gamma(u) \cap \Gamma(v)|}{|\Gamma(u) \cup \Gamma(v)|}$

Adamic/Adar:

The Adamic/Adar (Adamic Lada and Adar 2003) metric is defined as $AA(u, v) = \sum_{n \in \Gamma(u) \cap \Gamma(v)} \frac{1}{{\rm log}|\Gamma(n)|}$, where n ∈ N, N = Γ(u) ∩ Γ(v) is the set of common neighbors of u and v

Preferential Attachment:

The preferential attachment (Barabasi et al. 2002) metric is the multiplication of nodes u and v's degrees, PA(u, v) = Γ(u).Γ(v)

Katz:

Leo Katz proposed this metric in katz (1953); Katz metric sums all paths that exist between nodes u and vand penalizes the...