Heuristic for Network Coverage Optimization Applied in Finding Organizational Change Agents

Authors compare different ways of selecting change agents within network analysis paradigm and propose a new algorithm of doing so. All methods are evaluated against network coverage measure that calculates how many network members can be directly reached by selected nodes. Results from the analysis of organizational network show that compared to other methods the proposed algorithm provides better network coverage, at the same time selecting change agents that are well connected, influential opinion leaders.

💡 Research Summary

The paper addresses the problem of selecting change agents within an organization by focusing on the concept of network coverage—the proportion of employees that can be reached directly by a set of selected individuals. While traditional Social Network Analysis (SNA) relies on centrality measures such as degree, betweenness, closeness, and eigenvector to identify influential actors, these metrics do not explicitly optimize for direct reach.

To fill this gap, the authors propose a greedy heuristic algorithm that iteratively adds the node which contributes the largest number of previously uncovered nodes to the selected set. The algorithm operates on in‑degree (the number of incoming ties) for two reasons: (1) in‑degree reflects recognition by peers, offering a more objective assessment of authority, and (2) it mitigates data‑sparsity problems common in survey‑based network data because a missing response can be inferred from others’ nominations.

Algorithm steps

1. Compute in‑degree for every vertex and sort nodes in descending order.
2. While the cumulative coverage is below a predefined threshold (e.g., 80 % of the network), scan the sorted list of unselected nodes, calculate how many new nodes each would cover (including itself), and select the one with the greatest marginal gain.
3. Update the overall coverage and repeat.

The problem is analogous to the NP‑hard minimum set‑cover or knapsack problem; the greedy approach does not guarantee a global optimum but offers a tractable solution for real‑world networks.

Empirical evaluation
The method was tested on an intra‑organizational network of a production‑sales company comprising 215 employees. Data were collected via a sociometric questionnaire asking respondents to name colleagues they work with daily; 142 employees responded (66 % response rate). The resulting directed network had 2,225 edges, a density of 4.8 %, and an average degree of 20.69.

The authors compared five selection strategies: (i) the proposed greedy algorithm, (ii) ranking by in‑degree, (iii) betweenness, (iv) closeness, and (v) eigenvector centrality. Coverage was measured as the percentage of all employees that are directly reachable from the selected nodes. When 50 nodes (≈23 % of the workforce) were chosen, the greedy algorithm achieved 83 % coverage, outperforming the other methods (in‑degree 81 %, betweenness 80 %, closeness 79 %, eigenvector 77 %). The Pareto 80/20 principle held across all methods: roughly 20 % of the nodes accounted for about 80 % of the network reach.

To assess the quality of the selected nodes beyond raw coverage, the authors computed Spearman rank‑order correlations between the algorithmic ranking and each centrality ranking. Correlations were high and statistically significant (p < 0.01): in‑degree 0.921, betweenness 0.834, closeness 0.782, eigenvector 0.689. This indicates that the greedy set not only maximizes direct reach but also consists of individuals who are central in the traditional SNA sense (authority, bridging, and influence).

Strengths and contributions

• Introduces network coverage as an explicit optimization goal for change‑agent selection.
• Provides a simple, computationally cheap heuristic that outperforms standard centrality‑based selections in direct reach.
• Demonstrates that the resulting set aligns well with established notions of influence, bridging the gap between coverage and centrality.

Limitations and future work

1. The study is confined to a single medium‑sized organization; generalizability to larger or more hierarchical networks remains untested.
2. The greedy algorithm may miss the global optimum; exploring meta‑heuristics (e.g., genetic algorithms, simulated annealing) could yield better solutions.
3. Coverage is measured only at the one‑step level; real diffusion processes often involve multi‑step propagation, suggesting the need for models that incorporate cascade dynamics.
4. The current framework does not consider cost constraints (e.g., training or incentivizing change agents), which could be integrated into a multi‑objective optimization.

Conclusion
The paper presents a pragmatic approach to identifying organizational change agents by maximizing the proportion of employees they can directly contact. The proposed greedy in‑degree‑based heuristic consistently outperforms traditional centrality‑based rankings in coverage while still selecting highly central individuals. This dual advantage makes the method valuable for practitioners seeking to deploy a small, effective team of change agents. Future research should test the approach across diverse organizational contexts, incorporate multi‑step diffusion models, and explore cost‑aware, multi‑objective extensions.