Skip to content
Back to search results

Weighted Closeness Centrality

Introduced by
Opsahl, T.; Agneessens, F.; Skvoretz, J.
Year
2010
Type
Centrality measure (network analysis)
Family
Network science
Font size

What does it do?

Weighted closeness centrality extends the classic closeness measure to networks where edges carry numerical weights — such as frequency, strength, or cost — by incorporating those weights into shortest-path distances.

Nodes that can reach others quickly along strong or efficient connections receive higher scores, making it a richer indicator of information-spreading potential than its binary counterpart.

The intuition behind the method

In an unweighted network, closeness centrality asks how many steps a node needs to reach everyone else. In a weighted network that logic ignores whether ties are strong or weak. Weighted closeness treats each edge weight as a capacity or inverse distance: stronger ties count as shorter hops. A node scoring high is not just well-connected in the topological sense — it is genuinely close to others through high-capacity or low-cost paths. Opsahl et al. (2010) introduced a tuning parameter alpha that lets the analyst emphasize either path length or edge weight, bridging unweighted and purely weight-based closeness.

History

Classical closeness centrality was formalized by Sabidussi (1966) and later popularized by Freeman (1978). The extension to weighted networks was systematically developed by Opsahl, Agneessens, and Skvoretz in a 2010 Social Networks paper that generalized both degree and shortest-path measures using a tuning parameter alpha. Their framework became the standard reference for weighted centrality in social network analysis and has been implemented in widely used packages including igraph and NetworkX.

When to use it

Use weighted closeness centrality when edge weights (collaboration frequency, communication volume, monetary flow, transportation capacity) are theoretically meaningful and should influence which nodes are considered central. It is appropriate for social networks, trade networks, transportation systems, co-authorship graphs, and any setting where reach efficiency depends on tie strength. Prefer it over unweighted closeness whenever ignoring weights would distort the centrality ranking. Do not use it when weights are unreliable, arbitrarily scaled, or absent; in disconnected graphs without careful handling of unreachable pairs; or when the research question concerns pure structural topology rather than flow efficiency.

Limitations and considerations

  • Requires complete, reliable weight data; missing or noisy edge weights propagate directly into centrality scores.
  • Disconnected networks require special treatment (e.g., excluding unreachable pairs or using harmonic closeness).
  • Choosing the alpha parameter requires theoretical justification — arbitrary defaults can distort rankings.
  • Computationally more expensive than degree centrality for large sparse networks.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Incorporates edge weights so that strong or efficient connections properly boost a node's centrality score.
  • The alpha tuning parameter provides a continuous bridge between binary and fully weight-based closeness.
  • Retains the intuitive interpretation of closeness: how quickly a node can reach or be reached by others.
  • Directly comparable to unweighted closeness, enabling controlled sensitivity analyses.
  • Well-suited to directed weighted networks when combined with in-closeness and out-closeness decomposition.

Applications

  • Identifying key brokers in scientific collaboration networks based on co-authorship frequency.
  • Ranking airports or logistics hubs by reach efficiency in weighted transportation or trade networks.
  • Detecting influential spreaders in weighted communication or social media networks.
  • Evaluating information diffusion potential in weighted organizational or knowledge-sharing networks.
  • Comparing centrality rankings under different alpha values to assess robustness of findings.
  • Complementing weighted betweenness centrality to distinguish efficient reachability from flow control.

How it works

How it works

1

Edge weights are converted to distances: for collaboration or frequency networks the weight w is inverted (distance = 1/w) so that stronger ties yield shorter distances. For cost or latency networks the weight is already a distance. Dijkstra's algorithm then finds the minimum weighted distance d^w(i,j) between every pair of nodes.

Formula
d^w(i,j) = \min_{\text{paths}} \sum_{e \in \text{path}} w_e^{-1}
2

For each node i, the weighted distances to all other n-1 reachable nodes are summed. Nodes that are only weakly connected — or disconnected — to large parts of the network accumulate large sums, penalizing their centrality score.

3

The weighted closeness of node i is (n-1) divided by the sum of weighted distances to all other nodes, the same algebraic form as classic closeness but with d^w replacing hop counts. This normalizes the score to the interval (0, 1] so networks of different sizes remain comparable.

Formula
C^w(i) = \frac{n-1}{\sum_{j \neq i} d^w(i,j)}
4

Opsahl et al. introduced a parameter alpha (0 to 1) that blends pure topological distance (alpha = 0, equivalent to unweighted closeness) with pure weight information (alpha = 1). Intermediate values allow the researcher to control how much tie strength matters relative to path structure, making the measure adaptable to diverse empirical settings.

Formula
C^{w,\alpha}(i) = \frac{n-1}{\sum_{j \neq i} d^{w,\alpha}(i,j)}

Common pitfalls

  • Using raw counts as weights without checking for the appropriate weight-to-distance transformation (e.g., inverting frequency weights).
  • Applying the measure to a disconnected graph without adjusting for unreachable pairs, which inflates sum-of-distances for isolated nodes.
  • Treating centrality rankings as stable without testing sensitivity to the alpha parameter.
  • Conflating weighted closeness with betweenness centrality — closeness reflects reach efficiency, not flow control.
  • Ignoring normalization when comparing nodes across networks of different sizes.

Frequently asked questions

How should I choose the alpha parameter?

Alpha = 0 recovers unweighted closeness (topology only); alpha = 1 makes weights dominate. A common approach is to compute rankings at alpha = 0, 0.5, and 1 and report the sensitivity. If theory suggests tie strength is the primary mechanism, alpha = 1 is defensible; if topology matters more, use alpha = 0 or 0.5.

What weight-to-distance transformation should I use?

For frequency, strength, or capacity weights (higher = stronger), invert the weight: distance = 1/w. For cost, latency, or resistance weights (higher = weaker), use the weight directly as distance. Mixing these conventions without care will invert the centrality rankings.

How do I handle disconnected components?

The standard approach is to restrict the sum to reachable nodes and normalize by (n_reachable - 1) rather than (n - 1), or to use harmonic closeness centrality which naturally handles unreachable pairs by summing the reciprocal of distances.

Is weighted closeness better than unweighted closeness?

When weights carry substantive meaning (e.g., frequency of contact, volume of trade), weighted closeness gives a more accurate picture of reach efficiency. When weights are unreliable or weights and topology are nearly uncorrelated, the added complexity may not improve inference.

How does this differ from weighted betweenness centrality?

Weighted closeness measures how efficiently a node can reach others through strong paths. Weighted betweenness measures how often a node lies on the shortest paths between other pairs. A node can be close (high closeness) without being a critical bridge (high betweenness), and vice versa.

Related methods

Sources

  1. Opsahl, T., Agneessens, F. & Skvoretz, J. (2010). Node centrality in weighted networks: Generalizing degree and shortest paths. Social Networks, 32(3), 245–251.

    Open DOI reference
    Copied.
  2. Brandes, U. (2001). A faster algorithm for betweenness centrality. Journal of Mathematical Sociology, 25(2), 163–177.

    Open DOI reference
    Copied.