一句话：通过访谈识别依赖废弃影响性的四类情境因素，并构建基于 LLM 的分类器预测其是否重要。
开源依赖被废弃带来风险，但现有自动检测无法判断哪些废弃真正影响下游项目。作者访谈22位开源维护者，凝练出四类情境因素：集成深度、替代方案可用性、功能重要性、外部环境压力，并据此提出自动化通知工具的信息需求与设计需求。随后构建基于 LLM 的分类器，预测某依赖废弃在特定使用情境下是否会产生影响，并用124位维护者的独立用户研究评估。结果表明分类器能有效预测废弃的影响性，且 LLM 生成的理论解释对许多开发者判断废弃影响有帮助。
Despite the inevitable risk posed by abandoned open source dependencies, many developers feel they lack sufficient guidance on how to deal with dependency abandonment. Automated detection of abandonment is feasible, but not all abandoned dependencies impact downstream projects equally. In this paper, we perform a need-finding interview study with 22 open source maintainers exploring what makes the abandonment of certain dependencies impactful to their project given the context of their usage, as well as their information needs and design requirements for such an automated notification tool. We identify four categories of context-specific information that often affect the impactfulness of a given dependency’s abandonment to a particular downstream project: the depth of integration, the availability of alternatives, the importance of the functionality, and external environmental pressures. Using this emerging theory, we then build an LLM-based classifier to predict the impact of a dependency’s abandonment in a given context, and evaluate it with an independent user study with 124 open source maintainers. Our results show that the classifier is effective at predicting whether a dependency’s abandonment would be impactful to a project and that LLM-generated theory-based explanations are useful to many developers when making judgments about the potential impactfulness of abandonment.
@inproceedings{paperbot11557,
title = {Designing Abandabot: When Does Open Source Dependency Abandonment Matter?},
author = {Courtney Miller and Hao He and Weigen Chen and Elizabeth Lin and Chenyang Yang and Bogdan Vasilescu and Christian Kästner},
booktitle = {ICSE},
year = {2026},
doi = {10.1145/3744916.3764519}
}