Recommendation Systems for Evolving Software

在软件生命周期中，维护/演化阶段被认为是最长和最昂贵的。随着每次迭代，软件项目变得更大、更复杂，文档往往不完整。维护人员和重新设计人员需要使用大量数据的工具来提供及时和相关的建议。在更广泛的软件工程领域，软件工程推荐系统（RSSE）已经成为这样的工具。该计划旨在揭示有关软件项目演变的见解，并利用这些见解提出演化软件推荐系统（RSES）。 在软件项目上执行的更改具有不同的范围和动机，这些范围和动机可能与以后的几个版本相关。这些演化决策不一定有文档记录，范围从简单的重命名到词汇变化或库的替换。违反这些决定最终会导致严重的问题：从一致性和理解问题到恶性缺陷。就像“代码气味”被用于可能隐藏的实现问题的症状一样，违反先前的进化决策被称为“进化气味”，即，可能的修改问题的症状。该计划涵盖两个主要轴：（i）技术，有助于揭示变化和进化决策和（ii）建议有效的RSES的基础上，进化气味的调查。我们一直致力于第一个轴，我们（和其他人）开发的通用技术已经足够成熟，可以作为初步探索进化气味的工具。因此，第二个轴将是直接的焦点，并将通知随后的研究工作，以改善技术有关检索准确和有意义的软件演化数据。 本五年周期的目标如下： 1)收集/研究变化的事实和调查进化的气味 2)为不断发展的软件提出推荐系统 3)改进挖掘进化决策的技术 4)改进软件演化中的图差分方法。 第一个目标将通过系统的文献综述和研究将用于研究进化气味的项目语料库来实现。第二个项目将收集和研究关于区域社会经济系统的工具和出版物，以便提出一个通用架构和区域社会经济系统的第一个迭代。最后两个目标是关于改进技术，可以（i）以有见地的方式聚合更改事实或（ii）计算软件的两个版本之间的差异。 该计划研究有助于工程师完成任务的技术，从而使软件进化研究更接近工业相关性。强大和积极的学生将受到培训，并将有助于在发现和利用有关软件进化的新见解方面取得重大进展

Within the software life cycle, the maintenance/evolution phase is viewed as the longest and costliest. Software projects become bigger and more complex with each iteration and documentation is often incomplete. Maintainers and re-engineers need tools that use the sheer amount of data to make timely and relevant suggestions. In the broader field of software engineering, Recommendation Systems in Software Engineering (RSSE) have emerged as such tools. The proposed program aims at uncovering insights about a software project's evolution and using those to propose Recommendation Systems for Evolving Software (RSES). Changes performed on a software project have different scopes and motivations that can be relevant several versions later. These evolution decisions are not necessarily documented and range from simple renamings to vocabulary shifts or replacements of libraries. Violations of such decisions can eventually result in serious issues: from consistency and comprehension problems to vicious defects. Much like "code smells" is used for symptoms of possible hidden implementation problems, violations of previous evolution decisions qualify as “evolution smells”, i.e., symptoms of possible modification problems. This program covers two main axes: (i) techniques that help in uncovering changes and evolution decisions and (ii) proposals for effective RSES based on the investigation of evolution smells. We have been working on the first axis and the generic techniques we (and others) have developed are mature enough to serve as tools for preliminary explorations of evolution smells. The second axis will thus be the immediate focus and will inform the subsequent research efforts for improving techniques concerned with retrieving accurate and meaningful software evolution data. For this five-year cycle, the objectives are as follows: 1) Collect/study change facts and investigate evolution smells 2) Propose recommendation systems for evolving software 3) Improve techniques for mining evolution decisions 4) Improve approaches for diagram differencing in software evolution. The first objective will be achieved through a systematic literature review and the study of a corpus of projects that will be used to investigate evolution smells. The second will collect and study tools and publications on RSSEs in order to propose a generic architecture and first iterations of RSES. The last two objectives are about improving techniques that can (i) aggregate change facts in an insightful way or (ii) compute the differences between two versions of a software. This program investigates technologies that will help engineers in their tasks and thus brings research on software evolution closer to industrial relevance. Strong and motivated students will be trained and will help contribute to significant advances in the discovery and exploitation of new insights about software evolution