Developer chatrooms (e.g., the Gitter platform) are gaining popularity as a communication channel among developers. In developer chatrooms, a developer ( asker ) posts questions and other developers ( respondents ) respond to the posted questions. The interaction between askers and respondents results in a discussion thread . Recent studies show that developers use chatrooms to inquire about issues, discuss development ideas, and help each other. However, prior work focuses mainly on analyzing individual messages of a chatroom without analyzing the discussion thread in a chatroom. Developer chatroom discussions are context-sensitive, entangled, and include multiple participants that make it hard to accurately identify threads. Therefore, prior work has limited capability to show the interactions among developers within a chatroom by analyzing only individual messages. In this article, we perform an in-depth analysis of the Gitter platform (i.e., developer chatrooms) by analyzing 6,605,248 messages of 709 chatrooms. To analyze the characteristics of the posted questions and the impact on the response behavior (e.g., whether the posted questions get responses), we propose an approach that identifies discussion threads in chatrooms with high precision (i.e., 0.81 F-score). Our results show that inactive members responded more often and unique questions take longer discussion time than simple questions. We also find that clear and concise questions are more likely to be responded to than poorly written questions. We further manually analyze a randomly selected sample of 384 threads to examine how respondents resolve the raised questions. We observe that more than 80% of the studied threads are resolved. Advanced-level/beginner-level questions along with the edited questions are the mostly resolved questions. Our results can help the project maintainers understand the nature of the discussion threads (e.g., the topic trends). Project maintainers can also benefit from our thread identification approach to spot the common repeated threads and use these threads as frequently asked questions (FAQs) to improve the documentation of their projects.

中文翻译：

---

Gitter平台开发者讨论的实证研究

开发人员聊天室（例如，Gitter 平台）作为开发人员之间的沟通渠道越来越受欢迎。在开发者聊天室中，开发者 (提问者) 发布问题和其他开发人员 (受访者) 回复发布的问题。提问者和回答者之间的互动导致讨论线. 最近的研究表明，开发人员使用聊天室来询问问题、讨论开发想法并互相帮助。然而，先前的工作主要集中在分析聊天室的单个消息，而不分析聊天室中的讨论线程。开发人员聊天室讨论是上下文相关的、纠缠不清的，并且包含多个参与者，因此很难准确识别线程。因此，先前的工作仅通过分析单个消息来显示聊天室中开发人员之间的交互的能力有限。在本文中，我们通过分析 709 个聊天室的 6,605,248 条消息，对 Gitter 平台（即开发者聊天室）进行了深入分析。分析发布问题的特征以及对响应行为的影响（例如，发布的问题是否得到响应），我们提出了一种方法，可以高精度地识别聊天室中的讨论线程（即 0.81 F 分数）。我们的结果表明，与简单问题相比，不活跃成员的回复频率更高，独特的问题需要更长的讨论时间。我们还发现，与写得不好的问题相比，清晰简洁的问题更有可能得到回应。我们进一步手动分析随机选择的 384 个线程样本，以检查受访者如何解决提出的问题。我们观察到超过 80% 的研究线程得到解决。高级/初学者级别的问题以及已编辑的问题是最常解决的问题。我们的结果可以帮助项目维护者了解讨论线程的性质（例如，主题趋势）。