In a graph, two vertices with many common neighbours often occupy similar structural positions. The common-neighbours score exploits triadic closure: if A connects to B and C, and B connects to C, the open triad may close into a triangle. This pattern reflects homophily, community structure, social transitivity, and latent proximity. Under suitable generative models, shared neighbours increase the conditional probability that an unobserved edge exists, especially in sparse, clustered networks. The signal is not universal. Degree heterogeneity can create high scores through popularity alone, while bipartite structure, disassortativity, temporal growth, missing-not-at-random observations, and resolution limits can reverse the association. Common-neighbours methods therefore compete with preferential attachment, Adamic–Adar, resource-allocation, Katz, matrix factorisation, and graph neural networks. Evaluation requires temporal or inductive splits, negative-sampling controls, calibrated probabilities, and metrics such as average precision or precision at k; naïve random splits often inflate apparent significance through leakage.
“Two nodes with many shared neighbours are often close enough in a network that the missing edge is not random noise but an unobserved relation. ”
Reflect
When a connection is invisible, what surrounding pattern would persuade you that it is still there?
Visual Trail
A guided visual explanation assembled from QE artwork and sourced documentary images.
QE visual interpretationFrame 01
Shared neighbours predict a missing edge because they reveal triadic closure, community structure, and latent proximity—but degree and sampling bias can mislead.
Source: AI-generated visual interpretation
Creator: Question Everything
Limitation: This image explains or evokes the subject. It is not documentary evidence and should not be used to verify a factual claim.
Evidence
Verified claims with confidence scoring and cited sources.
1 of 3 findings need extra caution. Finding 1 rests on weaker sourcing than the other findings.
Claims remain in the reading flow. Select a citation number to inspect the source behind it.
In a network snapshot, two unconnected nodes may occupy overlapping local neighbourhoods. Common neighbours then act as structural evidence of proximity: co-authors who share collaborators, for example, may be more likely to collaborate later. Liben-Nowell and Kleinberg formalised this link-prediction problem and found that several proximity measures performed 40 to 50 times better than chance on their data, though exogenous events remain difficult to infer from topology alone.
A common-neighbour score counts, or otherwise weights, the vertices adjacent to both candidate endpoints. The intuition is local homophily: nodes embedded in partly shared social or functional contexts have greater structural affinity. CCPA extends this family by combining common-neighbour information with centrality, while evaluation across eight standard data sets reported improved predictability relative to established algorithms. Such scores remain correlational; they do not identify the social mechanism that creates an edge.
Traditional predictors include common neighbours, Adamic–Adar, preferential attachment and Katz index. Graph neural networks instead learn representations by aggregating information from graph structure and node or edge attributes, with node-based and subgraph-based approaches offering different expressive power. This can improve on a single local heuristic when relevant features exist, but incomplete graphs, hidden confounding and distribution shift still limit interpretation and out-of-sample reliability.
The complete record below preserves every citation, confidence input and recorded limitation.
In a network snapshot, two unconnected nodes may occupy overlapping local neighbourhoods. Common neighbours then act as structural evidence of proximity: co-authors who share collaborators, for example, may be more likely to collaborate later. Liben-Nowell and Kleinberg formalised this link-prediction problem and found that several proximity measures performed 40 to 50 times better than chance on their data, though exogenous events remain difficult to infer from topology alone.
Scored as if sourced, but every citation failed verification.
NO SURVIVING CITATION
Supporting passage
In a network snapshot, two unconnected nodes may occupy overlapping local neighbourhoods. Common neighbours then act as structural evidence of proximity: co-authors who share collaborators, for example, may be more likely to collaborate later. Liben-Nowell and Kleinberg formalised this link-prediction problem and found that several proximity measures performed 40 to 50 times better than chance on their data, though exogenous events remain difficult to infer from topology alone.
Citations (0 of 1 survived verification)
Nothing openable. Every citation was removed by provenance validation.
What limits this
A common-neighbour score counts, or otherwise weights, the vertices adjacent to both candidate endpoints. The intuition is local homophily: nodes embedded in partly shared social or functional contexts have greater structural affinity. CCPA extends this family by combining common-neighbour information with centrality, while evaluation across eight standard data sets reported improved predictability relative to established algorithms. Such scores remain correlational; they do not identify the social mechanism that creates an edge.
One source, not peer-reviewed. Thinner than the score suggests.
Supporting passage
A common-neighbour score counts, or otherwise weights, the vertices adjacent to both candidate endpoints. The intuition is local homophily: nodes embedded in partly shared social or functional contexts have greater structural affinity. CCPA extends this family by combining common-neighbour information with centrality, while evaluation across eight standard data sets reported improved predictability relative to established algorithms. Such scores remain correlational; they do not identify the social mechanism that creates an edge.
Citations
What limits this
Traditional predictors include common neighbours, Adamic–Adar, preferential attachment and Katz index. Graph neural networks instead learn representations by aggregating information from graph structure and node or edge attributes, with node-based and subgraph-based approaches offering different expressive power. This can improve on a single local heuristic when relevant features exist, but incomplete graphs, hidden confounding and distribution shift still limit interpretation and out-of-sample reliability.
One source, not peer-reviewed. Thinner than the score suggests.
Supporting passage
Traditional predictors include common neighbours, Adamic–Adar, preferential attachment and Katz index. Graph neural networks instead learn representations by aggregating information from graph structure and node or edge attributes, with node-based and subgraph-based approaches offering different expressive power. This can improve on a single local heuristic when relevant features exist, but incomplete graphs, hidden confounding and distribution shift still limit interpretation and out-of-sample reliability.
Citations
What limits this
Interactive Exploration
These visualizations respond to your curiosity. Interact to go deeper.
Represent the graph
Find candidate pairs
Count shared neighbours
Weight and rank
Validate temporally
| Local overlap | Global or learned structure | |
|---|---|---|
| Common neighbours | Counts shared adjacent nodes | Sensitive to local clustering |
| Katz index | Uses paths of varying length | Captures broader connectivity |
| Graph neural network | Aggregates neighbourhood information | Can incorporate node and edge features |
Tap any row to highlight and compare
Visual Gallery
Historical images, diagrams, and visual knowledge from Wikimedia Commons.
Perspectives
Enter a viewpoint. Notice what it reveals, what it leaves out, and whether it changes the question for you.
The strongest scientific account is structural: shared neighbours encode latent proximity, triadic closure or membership in the same densely connected region. Yet this is not universal. Temporal ordering, degree heterogeneity, missing-not-at-random observations and external events can break the relationship. The central debate is therefore not whether common neighbours help on average, but when their marginal information exceeds that of preferential attachment, path-based scores, node attributes or learned subgraph representations.
What this lens notices
Media
Curated media selected for this investigation.
What do you think?
No AI help here — no suggestions, no autocomplete, nothing finishing your sentences. That is deliberate. Working out what you think is effortful, and the effort is the part that changes you: reasoning is trained like a muscle, and a muscle that is always carried gets weaker. Let something else do the thinking and you keep the answer but lose the capacity to have reached it.
Write your current position.
Sign in to leave a mark. Your draft is saved here in the meantime.