Triadic Closure and Weak Ties | Machine Learning and Real-World Data

Triadic Closure

Definition: if two people share a common friend, they are likely to become friends themselves over time. This is a fundamental generative principle of social networks and the primary mechanism producing high clustering coefficients.

If A is friends with B and A is friends with C, the triad $\{A, B, C\}$ is “open.” Over time, B and C are likely to connect through social forces: opportunity (A introduces them), trust (A vouches for both), and homophily (they are similar because they both know A). When B–C forms, the triad “closes.”

Each closed triad contributes to the clustering coefficient of all three nodes. In a network with high triadic closure, $C_i$ values are consistently high (0.3–0.8 in real friend networks), reflecting the dense local structure.

Weak Ties (Granovetter, 1973)

Definition: ties with low emotional intensity, infrequent contact, and low mutual commitment, but high information-bridging value.

Strong ties (close friends, family, regular collaborators) form dense, overlapping clusters because of triadic closure: all your close friends know each other. The information circulating within this group is largely redundant — everyone hears the same things.

Weak ties (acquaintances, former colleagues, friends-of-friends) connect you to separate social circles. These ties bridge between different dense clusters and provide access to novel information — news, opportunities, ideas — that does not circulate within your close-knit group.

“The strength of weak ties”: Granovetter found that job-seekers were more likely to find new jobs through weak ties (acquaintances) than through strong ties (close friends and family). Close friends share the same information pool; acquaintances provide access to entirely different pools. Weak ties are weak in emotional commitment but strong in informational value.

Local Bridges

Definition: an edge $(u, v)$ is a local bridge if its endpoints share no common neighbours. The neighbourhoods of $u$ and $v$ are entirely disjoint except for the edge $(u, v)$ itself.

Removing a local bridge forces the distance between $u$ and $v$ to become at least 3 (since no neighbour of $u$ is also a neighbour of $v$ ). However, the graph may remain connected through longer paths.

Distinction from a bridge (cut-edge): a bridge is an edge whose removal disconnects the graph. Every bridge is a local bridge (if removal disconnects, the distance becomes infinite, which is certainly > 2). But the converse fails: in the 2020 Q9 $4 \times 4$ grid, there are no bridges (removing any single edge leaves the grid connected via alternative routes around the perimeter). But there are local bridges: edges where the two nodes share no common neighbour — this happens at corners and edges of the grid.

In social networks: local bridges are almost always weak ties. A strong tie between two people who share no common friends is sociologically unlikely (triadic closure would have introduced a common friend). Therefore, when you find a local bridge in a social network, you have almost certainly found a weak tie — and a crucial conduit for novel information.

Connection to Edge Betweenness and Newman-Girvan

Weak ties acting as local bridges have exceptionally high edge betweenness centrality: they lie on the shortest paths between many pairs of nodes in different social clusters. If two dense communities are connected only by a few weak ties, nearly all shortest paths between the communities pass through those ties.

This makes them the primary targets in the Newman-Girvan community detection algorithm: repeatedly removing the highest-betweenness edges breaks the weak ties first, splitting the graph into its constituent communities. The algorithm effectively detects where the weak ties are by computing edge betweenness.

Summary

Concept	Definition	Significance
Triadic closure	Friends of friends become friends	Explains high clustering; fundamental generative principle
Strong ties	Close, frequent, emotionally intense	Form dense overlapping cliques; redundant information
Weak ties	Distant, infrequent, low commitment	Bridge between communities; conduit for novel information
Local bridge	Endpoints share no common neighbours	High edge betweenness; target for Newman-Girvan removal
Bridge	Removal disconnects the graph entirely	Rare in real networks; stronger condition than local bridge

Past paper questions: y2020p3q9