[Paper Review] Team Formation: Striking a Balance between Coverage and Cost

Alina Ene, Sofia Maria Nikolakaki+1 more|arXiv (Cornell University)|Feb 18, 2020

Mobile Crowdsensing and Crowdsourcing5 references4 citations

TL;DR

This paper introduces a novel team-formation model that balances skill coverage and team cost by optimizing a submodular objective minus linear costs. Using greedy algorithms with provable approximation guarantees, the approach achieves high efficiency and effectiveness on real crowdsourcing data, offering a generalizable framework for cost-coverage trade-offs in team selection.

ABSTRACT

Motivated by online crowdsourcing platforms as well as applications that span human-resource management in industrial and research organizations researchers have studied extensively the team-formation problem. This problem has been primarily formalized as an optimization problem where the goal is to optimize some metric of the team performance, subject to the constraint that the team members should cover all the skills required by the task. In this paper, we generalize this problem formulation and set as our objective to optimize a function that maximizes the coverage of skills minus the cost of the corresponding team. This formulation appears as a more natural one, particularly in cases where one needs to strike a balance between the coverage achieved and the cost being paid. To the best of our knowledge we are the first to formalize the team-formation problem in this manner. From the algorithmic perspective, we demonstrate that by using simple variants of the standard greedy algorithm (used for submodular optimization) we can design algorithms that have provable approximation guarantees, are extremely efficient and work very well in practice. Our experiments with real data from online crowdsourcing platforms demonstrate the efficiency and the efficacy of our methods. Finally, we believe that our problem formulation and algorithms are of independent interest and can be used in many applications where there is a submodular objective and a linear cost.

Motivation & Objective

• Motivate the need for a balanced approach between team coverage and cost in human-resource management and crowdsourcing platforms.
• Address the limitation of prior work that optimizes only for coverage or cost in isolation.
• Formalize team formation as a joint optimization of coverage minus cost, reflecting real-world trade-offs.
• Design efficient algorithms with theoretical guarantees for this new formulation.
• Demonstrate practical efficacy using real data from online crowdsourcing platforms.

Proposed method

• Formulate the team-formation problem as maximizing a submodular function (coverage) minus a linear cost function.
• Adapt standard greedy algorithms for submodular optimization to handle the combined objective.
• Prove approximation guarantees for the proposed greedy variants under the new objective.
• Ensure algorithmic efficiency by leveraging the submodularity and monotonicity properties of the coverage function.
• Implement and evaluate the algorithms on real crowdsourcing datasets to validate performance.
• Use the framework as a general tool for any application with submodular objectives and linear costs.

Experimental results

Research questions

• RQ1 How can team formation be formalized to balance skill coverage and team cost in a principled way?
• RQ2 Can greedy algorithms for submodular optimization be adapted to achieve theoretical guarantees under a combined coverage-cost objective?
• RQ3 How efficient and effective are the proposed algorithms in practice on real-world data?
• RQ4 What is the impact of varying cost thresholds on team composition and coverage?
• RQ5 In what broader applications can this formulation be usefully applied beyond team formation?

Key findings

• The proposed formulation of maximizing coverage minus cost is the first to explicitly model the trade-off between team quality and cost.
• The adapted greedy algorithms achieve provable approximation guarantees under the new objective function.
• The algorithms are highly efficient and scale well to large datasets, making them practical for real-world deployment.
• Experiments on real crowdsourcing data show strong performance in both coverage and cost efficiency.
• The framework is generalizable and applicable to any problem with submodular objectives and linear costs.
• The method outperforms baseline approaches that optimize only for coverage or cost in isolation.