Experts2team: Task Relevance-Induced Team Formation by Combining Global Cohesion with Local Decoupling

Yue Kou, Yingxuan Du, Derong Shen, Xiangmin Zhou, Dong Li, Tiezheng Nie, Ge Yu

Abstract: Effective teamwork is crucial for solving complex domain-specific problems, requiring expert teams tailored to task requirements. However, existing works overlook the diversity of expert collaboration types, tend to confine member selection within existing teams, and neglect the relevance between tasks. In this paper, we propose a task relevance-induced team formation framework by combining global cohesion with local decoupling (called Experts2team), ensuring the accuracy, diversity, and efficiency of team formation. Specifically, we first design a heterogeneous collaborative network (HCN) to more meticulously describe the collaborations among experts and propose a HCN-based method tailored for task-specific team division. Then we propose a comprehensive global-local approach to learning team representations, which effectively avoids the issue of scope limitation. We also propose a task relevance-induced team matching strategy to mitigate pseudo-failures during matching. We evaluate our method on four real-world datasets and the experimental results show the superiority of our method compared to state-of-the-art team formation methods.

External IDs:dblp:conf/dasfaa/KouDSZLNY25