While the domain of individual-level AI-assisted analysis has been extensively explored in previous studies, the field of AI-assisted collaborative qualitative analysis remains relatively unexplored. After identifying CQA practices and design opportunities through formative interviews, we introduce our collaborative qualitative coding tool, CoAIcoder, and designed the four different collaboration methods. We subsequently implemented a between-subject design involving 32 pairs of users who have undergone training in CQA across three commonly utilized phases under four methods. Our results suggest that CoAIcoder, which employs AI and a Shared Model, could potentially improve the efficiency of the coding process in CQA by fostering a quicker shared understanding and promoting early-stage discussions. However, this may come with the potential downside of reduced code diversity. We also underscored the existence of a trade-off between the level of independence and the coding outcome when humans collaborate during the early coding stages. Lastly, we identify design implications that could inspire and inform the future design of CQA systems.

翻译：尽管以往研究已在个体层面的AI辅助分析领域进行了广泛探索，但AI辅助协作定性分析这一领域仍相对未被充分研究。通过形成性访谈识别CQA实践与设计机会后，我们引入了协作定性编码工具CoAIcoder，并设计了四种不同的协作方法。随后，我们实施了组间设计实验，涉及32对已接受CQA训练的用户，在四种方法下完成三个常用阶段的任务。结果表明，采用AI与共享模型的CoAIcoder可能通过促进更快速的共享理解、鼓励早期阶段讨论来提升CQA编码过程的效率，但其潜在代价是降低编码多样性。同时，我们强调在早期编码阶段人类协作时，独立性与编码结果之间存在权衡关系。最后，我们提出了能启发并指导未来CQA系统设计的设计启示。