VaryMinions

From business processes to course management, variability-intensive software systems (VIS) are now ubiquitous. Such systems can configure their behaviours through the activation of options, e.g., to derive variants handling building permits across municipalities, or implementing different functionalities(quizzes, forums) for a given course. These customisation facilities allow VIS to support a variety of relevant customer requirements while taking advantage of reuse for common parts, thus realising both scope and scale economies.Behavioural differences amongst variants manifest themselves in event logs.Should an anomalous behaviour happen or a refactoring of the system needed, it is mandatory to know which variant(s) have produced this behaviour. Since variant information is barely present in logs, this paper supports this task by employing machine learning techniques to map behaviours (event sequences)to variants. Specifically, we train Long Short Term Memory (LSTMs) andGated Recurrent Units (GRUs), two kinds of recurrent neural networks, to relate event sequences with the variants they belong to on six different datasets issued from the configurable process and VIS domains. Our results, after having evaluated 20 different architectures of LSTM/GRU, demonstrate that it is possible to learn effectively the trace-to-variant mapping with a good accuracy(at least 80% and up to 99%) and at scale, i.e., identifying 50 variants using5000+ traces for each variant.从业务流程到课程管理领域，可变密集型软件系统（variability-intensive software systems，VIS）如今已无处不在。此类系统可通过激活可选配置项调整自身行为，例如生成可跨市政部门处理建筑许可的系统变体，或是为特定课程实现测验、论坛等差异化功能。这类定制化能力使得VIS既能复用通用模块，又能满足多样化的客户需求，从而实现范围经济与规模经济。不同系统变体间的行为差异会在事件日志中体现。若出现异常行为，或是需要对系统进行重构，我们必须明确究竟是哪些系统变体引发了该行为。由于事件日志中几乎未包含变体相关信息，本文借助机器学习技术，将行为（即事件序列）映射至对应的系统变体，以此解决该问题。具体而言，我们选取两类循环神经网络——长短期记忆网络（Long Short Term Memory，LSTM）与门控循环单元（Gated Recurrent Unit，GRU），基于来自可配置流程与VIS领域的6个不同数据集，训练模型将事件序列与其所属的系统变体进行关联。在评估了20种不同的LSTM/GRU架构后，我们的实验结果表明，该方法可高效学习事件轨迹到系统变体的映射关系，且准确率优异（最低可达80%，最高可达99%），同时具备良好的可扩展性：例如针对50个系统变体，每个变体使用5000条以上的事件轨迹即可完成识别。