Semantic Process Discovery from User Interaction Logs

流程挖掘广泛用于基于从IT系统提取的事件数据（存储在所谓的事件日志中）来发现、分析和改进业务流程。在这方面的一个关键任务是过程发现，其目的是重建过程是如何真正执行的。为此，流程发现努力在事件日志中捕获的记录行为的基础上建立准确的流程模型。然而，使用这样的事件日志作为发现的基础具有重要的限制：它将分析的范围限制为后端事件，即，由实际用户活动触发的次要间接事件。因此，不会导致此类后端事件或发生在生产力应用程序（如Excel和Outlook）中的用户活动不会记录在事件日志中，因此，传统的流程挖掘和发现技术无法看到这些活动。为了避免此问题并能够获得有关业务流程的全面视图，本提案的目标是启用基于用户交互（UI）日志的流程发现，而不是传统的事件日志。从本质上讲，UI日志是GUI组件上执行的记录交互的集合，例如单击按钮或文本区域中的键盘输入。使用用户界面日志的好处是，可以为在计算机上执行其活动的任何业务流程获取这些日志，而无需考虑其所需的特定应用程序。但是，从此类用户界面日志中获取流程信息是一项复杂的任务，需要克服各种问题。具体来说，我们需要解决两个问题领域，每个领域都有自己的具体挑战：1）数据转换和2）流程表示。数据转换的问题区域涉及将输入UI日志转换为事件日志，该事件日志携带信息性事件标签，没有噪声，并且具有适当的案例标识符。流程表示的问题领域涉及从日志中的低级别事件导出有用的表示。这需要将事件分组为更高级别的事件，为这些更高级别的活动生成适当的名称，并为用户提供在高级和低级信息之间平衡的有效过程表示。拟议的项目将通过将行为过程分析与新的语义角度相结合来解决这些挑战。拟议的项目将导致开发以自动化方式解决上述挑战的方法，最终涵盖从UI日志到信息流程表示的整个管道。

Process mining is widely used to discover, analyze, and improve business processes based on event data extracted from IT systems, stored in so-called event logs. A key task in this regard is process discovery, which aims to reconstruct how a process was truly executed. To do so, process discovery strives to establish an accurate process model on the basis of the recorded behavior captured in an event log. Using such event logs as basis for discovery has an important limitation, however: It limits the scope of analysis to back-end events, i.e., secondary, indirect events that were triggered by the actual user activity. User activities that do not result in such back-end events or take place in productivity applications such as Excel and Outlook, are thus not recorded in event logs and, therefore, invisible to traditional process mining and discovery techniques.To avoid this problem and be able to elicit a comprehensive view on business processes, the goal of this proposal is to enable process discovery based on user interaction (UI) logs, rather than on traditional event logs. In essence, a UI log is a collection of recorded interactions performed on GUI components, such as clicks on buttons or keyboard entries in text areas. The benefit of using UI logs is that they can be obtained for any business process of which the activities are performed on a computer, regardless of the specific applications required for it. However, eliciting process information from such UI logs is a complex task, for which various problems need to be overcome. Specifically, we need to address two problem areas, each with its own specific challenges: 1) data transformation and 2) process representation. The problem area of data transformation is concerned with turning the input UI log into an event log that carries informative event labels, is free of noise, and has a proper case identifier. The problem area of process representation is concerned with deriving a useful representation from the low-level events from the logs. This entails grouping events into higher-level events, generating proper names for these higher-level activities, and providing the user with an effective process representation that balances between high and low-level information. The proposed project will address these challenges by combining behavioral process analysis with a novel semantic angle. The proposed project will result in the development of approaches that address the aforementioned challenges in an automated manner, ultimately covering the entire pipeline from UI log to an informative process representation.