Risk-Aware Skill-Coverage Hybrid Workforce Configuration on Social Networks

Hui-Ju Hung, Guang-Siang Lee, Chia-Hsun Lu, Chih-Ya Shen, De-Nian YangarXiv 2026 · arXiv

TL;DR

This paper introduces a formal framework for optimizing hybrid work schedules by balancing professional collaboration against health and operational risks. The authors prove that finding the ideal onsite team is NP-hard and introduce GRIA, a highly effective three-phase algorithm that maximizes onsite collaboration while guaranteeing full skill coverage and keeping epidemic risks under a strict budget.

What Problem Does This Paper Address?

While hybrid work has become a staple of modern corporate life, deciding exactly who should work onsite on any given day remains a complex logistical challenge. Managers must ensure that location-specific tasks are covered by employees with the right skills, and they want to maximize the collaborative synergy of having people in the same room. At the same time, physical co-location introduces contact-based risks, such as the spread of seasonal influenza, COVID-19, or even the viral propagation of workplace rumors.

Existing research falls short by tackling these issues in isolation: team formation algorithms focus purely on skills and synergy, while epidemic-control models focus strictly on quarantines and containment without considering productivity. This paper bridges the gap by formulating the Risk-aware Skill-coverage Hybrid Workforce Configuration (RSHWC) problem, which models the workforce as a two-layer social network to simultaneously optimize skills, collaboration, and safety.

Fig. 1. Illustrative example of a two-layer social network

Methods

The authors model the workforce using a two-layer network: a Contact Network ($G_c$) that captures physical interactions and transmission probabilities, and a Partnership Network ($G_p$) that quantifies collaboration benefits both onsite and remotely.

To solve the NP-hard RSHWC problem, they propose the Guided Risk-aware Iterative Assembling (GRIA) algorithm, which operates in three distinct phases:

  1. Risk-aware Workforce Construction (RWC): An incremental phase that greedily adds employees to the onsite pool. It prioritizes candidates who maximize the ratio of marginal collaboration gain to contact risk ($\tau_{V^*}(v) / \text{risk}(v)$) until all required skills are covered.
  2. Skill-preserving Workforce Refinement (SWR): A pruning phase that filters out employees whose marginal collaboration contribution falls below the current team average, provided their removal doesn't break the required skill coverage.
  3. Risk-reducing Member Replacement (RMR): A local search phase that identifies high-risk onsite employees and attempts to swap them with safer, remote alternatives without sacrificing skill coverage or overall collaboration quality.

Datasets

The algorithm was evaluated on four diverse, real-world network datasets:

  • Manhattan & Virginia: Large-scale networks used to simulate realistic physical contact patterns.
  • ca-GrQc & ca-HepPh: Collaboration networks from arXiv (General Relativity and High Energy Physics) representing professional partnership structures.

Where skill or collaboration data was missing, the authors augmented the datasets using established network synthesis benchmarks from literature.

Key Insights and Ideas

  • The Two-Layer Elegance: Separating physical proximity (the contact layer) from professional synergy (the partnership layer) allows the model to capture real-world nuances—such as two close collaborators who work entirely digitally vs. two people who sit next to each other but rarely work together.
  • The Pruning Justification: The authors mathematically prove that removing below-average contributors from the onsite pool strictly improves or preserves the average collaboration score. This theoretical guarantee underpins the success of the SWR phase.
  • Risk vs. Reward Ratio: By evaluating candidates based on a collaboration-to-risk ratio rather than just raw skill or raw risk, the algorithm avoids the pitfall of blindly picking highly skilled individuals who happen to be super-spreaders in the contact network.

Key Results

  • Superior Collaboration: Across all four datasets, GRIA consistently achieved higher average collaboration scores than six state-of-the-art baselines (including MaxGF, ST-Exa, and Random Walk with Restart).
  • Robustness to Remote Effectiveness: The authors varied the effectiveness of remote work from 50% to 90% of onsite efficiency. GRIA maintained its performance edge across this entire spectrum, proving it adaptable to companies with highly mature remote-work infrastructures.
  • Computational Efficiency: While slower than purely structural baselines due to the overhead of calculating non-linear epidemic propagation risks, GRIA scale well ($O(|V|^2 \cdot c_r)$) and runs substantially faster than complex random-walk alternatives.

Limitations and Future Work

The current formulation assumes a static snapshot of both collaboration and risk networks. In reality, workplace interactions and epidemic states change over time (e.g., project lifecycles shift, or a flu wave peaks). The authors note that extending this framework to handle temporal dynamics and evolving propagation models is a promising direction for future research.

Summary updated

Abstract

In hybrid workforce configurations, it is important to decide which employees should work onsite or remotely while ensuring the collaboration benefits against contact-based health risks and skill requirements. In this paper, we formulate the Risk-aware Skill-coverage Hybrid Workforce Configuration (RSHWC) problem on a two-layer social network that balances physical contact risks and social collaboration ties to meet skill requirements. We prove that RSHWC is NP-hard and propose the Guided Risk-aware Iterative Assembling (GRIA) algorithm, a multi-stage algorithm that combines risk-aware workforce construction, skill-preserving workforce refinement, and risk-reducing member replacement. Experiments on four real-world networks show that GRIA consistently outperforms state-of-the-art baselines under various settings.
BibTeX
@misc{hung2026risk,
  author = {Hui-Ju Hung and Guang-Siang Lee and Chia-Hsun Lu and Chih-Ya Shen and De-Nian Yang},
  title = {{Risk-Aware Skill-Coverage Hybrid Workforce Configuration on Social Networks}},
  year = {2026},
  eprint = {2603.00727},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2603.00727},
}
Figure