While individual-level AI-assisted analysis has been fairly examined in prior work, AI-assisted collaborative qualitative analysis (CQA) remains an under-explored area of research. After identifying CQA behaviors and design opportunities through interviews, we propose our collaborative qualitative coding tool, CoAIcoder, and present the results of our studies that examine how AI-assisted CQA can work. We then designed a between-subject experiment with 32 pairs of novice users to perform CQA across three commonly practiced phases under four collaboration methods. Our results show that CoAIcoder (with AI & Shared Model) could potentially improve the coding efficiency of CQA, however, with a potential risk of decreasing the code diversity. We also highlight the relationship between independence level and coding outcome, as well as the trade-off between, on the one hand, Coding Time & IRR, and on the other hand Code Diversity. We lastly identified design implications to inspire the future design of CQA systems.

翻译：尽管个体层面的AI辅助分析已在先前工作中得到充分研究，但AI辅助协作定性分析（CQA）仍是一个未被充分探索的研究领域。通过访谈识别出CQA行为与设计机遇后，我们提出了协作定性编码工具CoAIcoder，并展示了探究AI辅助CQA可行性的研究成果。我们进一步设计了一项受试者间实验，招募32对新手用户在四种协作方法下完成CQA的三个常见实践阶段。研究结果表明，CoAIcoder（配备AI与共享模型）虽可能提升CQA的编码效率，但也存在降低编码多样性的潜在风险。我们同时揭示了独立性水平与编码结果之间的关联，以及编码时间与IRR（编码者间信度）同编码多样性之间的权衡关系。最终，我们提炼出设计启示，为未来CQA系统的设计提供参考。