跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2611.20225 · v1

Disparity in Influence: A Fairness Evaluation of Centrality Measures in Complex Networks

首次提交 2026-11-16 03:47
版本更新 2026-11-16 03:47
雷达观测 2026-11-17 12:06
arXiv ↗PDF ↗

摘要

Social networks are shaped by homophily-driven connections, which naturally lead to the emergence of communities with uneven sizes and influence. Such structural imbalances impact the fairness of network analysis tools, particularly centrality measures that are widely used to identify influential nodes for tasks like information diffusion and network robustness. In this work, we conduct a comprehensive assessment of fairness in centrality measures, using influence spreading across networks with heterogeneous community structures. We first analytically show that disparity is bounded by outreach; cross-community outreach is bounded by the inter-community cut; and when all seeds lie in one community, disparity approaches the normalized outreach within that community as cross-community transmission vanishes. This explains why attenuation concentrates influence in the community where diffusion starts. We further evaluate a broad spectrum of centrality measures, including local, semi-local, global, and community-aware measures, on both synthetic network generative models and real-world datasets, using the Independent Cascade and Linear Threshold Models. The results consistently show that influence tends to concentrate within larger communities, with disparities becoming more pronounced under constrained or deterministic diffusion settings. We further observe that fairness is highly sensitive to the choice of centrality measure rather than its general category. In particular, fairness-aware adaptations of PageRank demonstrate a strong balance between maximizing influence reach and maintaining equitable distribution across communities. On real networks, these conclusions are largely stable under random removal or addition of up to 3% of the edges. Our findings highlight the need to explicitly account for fairness when designing and selecting centrality-based strategies in networked systems.

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

目前只收录了一个版本。收录下一版后,可以对照标题、作者、分类和摘要。

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v1 · 2026-11-16 03:47 · 最新观测版本

Disparity in Influence: A Fairness Evaluation of Centrality Measures in Complex Networks

Social networks are shaped by homophily-driven connections, which naturally lead to the emergence of communities with uneven sizes and influence. Such structural imbalances impact the fairness of network analysis tools, particularly centrality measures that are widely used to identify influential nodes for tasks like information diffusion and network robustness. In this work, we conduct a comprehensive assessment of fairness in centrality measures, using influence spreading across networks with heterogeneous community structures. We first analytically show that disparity is bounded by outreach; cross-community outreach is bounded by the inter-community cut; and when all seeds lie in one community, disparity approaches the normalized outreach within that community as cross-community transmission vanishes. This explains why attenuation concentrates influence in the community where diffusion starts. We further evaluate a broad spectrum of centrality measures, including local, semi-local, global, and community-aware measures, on both synthetic network generative models and real-world datasets, using the Independent Cascade and Linear Threshold Models. The results consistently show that influence tends to concentrate within larger communities, with disparities becoming more pronounced under constrained or deterministic diffusion settings. We further observe that fairness is highly sensitive to the choice of centrality measure rather than its general category. In particular, fairness-aware adaptations of PageRank demonstrate a strong balance between maximizing influence reach and maintaining equitable distribution across communities. On real networks, these conclusions are largely stable under random removal or addition of up to 3% of the edges. Our findings highlight the need to explicitly account for fairness when designing and selecting centrality-based strategies in networked systems.

查看此版本解读 →查看此版本原文 ↗