Abstract
LLM-based agents excel at writing articles, coding, and information retrieval. However, they struggle to form strong collaborations within the scientific community due to the bidirectional, dynamic nature of the problem and the high demand for decision interpretability. We proposed COWEAVER, a bi-directional, learnable, and explainable algorithm to match scientists and foster strong collaborations within a human-agent network.
COWEAVER matches candidates and requesters by filling capability gaps and filters candidates through a two-stage ranking step.
Finally, the model explores newcomers by maintaining uncertainty-aware capability estimates and updating them through requester's feedback. We show that COWEAVER's selection mechanism, combining both exploration (UCB) and greedy strategies, exceeds the greedy-only mechanism—the analytical best solution—on 6 out of the 20 tasks and performed on par with the greedy-only mechanism in terms of selecting the best candidate.
We compared COWEAVER baselines in terms of matching quality and efficiency, demonstrating that COWEAVER outperforms baselines on all metrics.
Blogger's Review: COWEAVER's innovative bidirectional matching mechanism and decision process interpretability significantly enhance the efficiency and transparency of collaborations between scientists and agents. This research lays a vital theoretical foundation and practical guidance for future human-agent collaboration.