Graduate School of Management of Technology, Sogang University, 35 Baekbeom-ro, Mapo-gu, Seoul 04107, Korea.
Department of Computer Science and Engineering, Sogang University, 35 Baekbeom-ro, Mapo-gu, Seoul 04107, Korea.
Sensors (Basel). 2018 Sep 5;18(9):2963. doi: 10.3390/s18092963.
Starting with the Internet of Things (IoT), new forms of system operation concepts have emerged to provide creative services through collaborations among autonomic devices. Following these paradigmatic changes, the ability of each participating system to automatically diagnose the degree of quality it is providing is inevitable. This paper proposed a method to automatically detect symptoms that hinder certain quality attributes. The method consisted of three steps: (1) extracting information from real usage logs and automatically generating an activity model from the captured information; (2) merging multiple user activity models into a single, representative model; and (3) detecting differences between the representative user activity model, and an expected activity model. The proposed method was implemented in a domain-independent framework, workable on the Android platform. Unlike other related works, we used quantitative evaluation results to show the benefits of applying the proposed method to five Android-based, open-source mobile applications. The evaluation results showed that the average precision rate for the automatic detection of symptoms was 70%, and the success rate for user implementation of usage scenarios demonstrated an improvement of around 21%, when the automatically detected symptoms were resolved.
从物联网（IoT）开始，新的系统操作概念形式已经出现，通过自治设备之间的协作来提供创新服务。随着这些范式的变化，每个参与系统自动诊断其提供的质量程度的能力是不可避免的。本文提出了一种自动检测阻碍某些质量属性的症状的方法。该方法包括三个步骤：（1）从实际使用日志中提取信息，并从捕获的信息中自动生成活动模型；（2）将多个用户活动模型合并为单个代表性模型；（3）检测代表性用户活动模型与预期活动模型之间的差异。所提出的方法是在独立于域的框架中实现的，可在 Android 平台上使用。与其他相关工作不同，我们使用定量评估结果来展示将所提出的方法应用于五个基于 Android 的开源移动应用程序的好处。评估结果表明，在自动检测症状时，平均精度率为 70%，当自动检测到的症状得到解决时，用户实现使用场景的成功率提高了约 21%。
