Title:SUMMIT: Scaffolding OSS Issue Discussion Through Summarization

Abstract:For Open Source Software (OSS) projects, discussions in Issue Tracking Systems (ITS) serve as a crucial collaboration mechanism for diverse stakeholders. However, these discussions can become lengthy and entangled, making it hard to find relevant information and make further contributions. In this work, we study the use of summarization to aid users in collaboratively making sense of OSS issue discussion threads. We reveal a complex picture of how summarization is used by issue users in practice as a strategy to help develop and manage their discussions. Grounded on the different objectives served by the summaries and the outcome of our formative study with OSS stakeholders, we identified a set of guidelines to inform the design of collaborative summarization tools for OSS issue discussions. We then developed SUMMIT, a tool that allows issue users to collectively construct summaries of different types of information discussed, as well as a set of comments representing continuous conversations within the thread. To alleviate the manual effort involved, SUMMIT uses techniques that automatically detect information types and summarize texts to facilitate the generation of these summaries. A lab user study indicates that, as the users of SUMMIT, OSS stakeholders adopted different strategies to acquire information on issue threads. Furthermore, different features of SUMMIT effectively lowered the perceived difficulty of locating information from issue threads and enabled the users to prioritize their effort. Overall, our findings demonstrated the potential of SUMMIT, and the corresponding design guidelines, in supporting users to acquire information from lengthy discussions in ITSs. Our work sheds light on key design considerations and features when exploring crowd-based and machine-learning-enabled instruments for asynchronous collaboration on complex tasks such as OSS development.