一句话：提出 KnowDetector，利用领域知识构建自动化图并通过图模式匹配检测 TAP 应用间干扰。
触发-动作编程（TAP）简化了 IoT 自动化，但应用间异常交互会导致冲突、死循环等干扰，带来安全与安全隐患。现有方法多局限于处理显式简单冲突，忽略上下文影响，且大规模应用下性能不足。论文设计 KnowDetector，引入领域知识模型 KnowIoT，以命令级 IoT 服务及其与上下文环境的关系刻画设备；据此构建包含事件、条件、动作、物理环境通道及语义关系边的自动化图，并将干扰检测抽象为图模式匹配，总结四类十种干扰模式。在三个真实数据集上评估，KnowDetector 的精度、召回率与 F-measure 均优于现有最优工具，并能以较低时间开销扩展至大量应用中的干扰检测。
Trigger-action programming (TAP) is a widely used development paradigm that simplifies the Internet of Things (loT) automation. However, the exceptional interactions between automation applications may result in interferences, such as conflicts and infinite loops, which cause undesirable consequences and even security and safety risks. While several techniques have been proposed to address this problem, they are often restricted in handling explicit and simple conflicts without considering contextual influences. In addition, they suffer from performance issues when applying to large-scale applications. To address these challenges, we design an effective and practical tool KnowDetector with comprehensive domain knowledge to detect application interferences. To detect application interferences, KnowDetector constructs an automation graph with 1) events, conditions, and actions from automation applications, 2) vertices representing physical environment channels, and 3) edges derived from potential semantic relations between the vertices. In order to make the graph extensively capture the interactions between automation applications, we propose a knowledge model named KnowloT that accurately characterizes loT devices with command-level loT services and the intricate relations between these services and the contextual environment. We abstract the interference detection into a graph pattern-matching problem and summarize ten application interference patterns of four types. Finally, KnowDetector can efficiently detect application interferences by searching for sub-graphs matching the patterns within the automation graph. We evaluated KnowDetector on three real-world datasets. The results demonstrated that it outperformed the other state-of-the-art tools with the highest precision, recall, and F-measure. In addition, KnowDetector is scalable to detect application interferences within a large number of applications with a minimal time overhead.
@inproceedings{WangCLWWH23,
title = {Detecting Smart Home Automation Application Interferences with Domain Knowledge},
author = {Tao Wang and Wei Chen and Liwei Liu and Guoquan Wu and Jun Wei and Tao Huang},
booktitle = {ASE},
year = {2023},
doi = {10.1109/ase56229.2023.00070}
}